Data intelligent hierarchical storage method and system based on multi-dimensional quantitative evaluation

Abstract

The invention discloses a data intelligent hierarchical storage method and system based on multi-dimensional quantitative evaluation, and relates to the technical field of computer data storage. The method is used for solving the problems that cold and hot data identification lags behind, migration strategies lack collaboration and system performance fluctuation regulation and control are insufficient in traditional hierarchical storage. The method comprises the following steps: collecting dynamic access features and static attribute features of a data object in real time, adaptively adjusting feature weights according to activeness fluctuation, and calculating a hierarchical migration coefficient; a business operation sequence co-occurrence relation is analyzed to construct a data association graph, and a collaborative migration constraint is established in combination with density clustering and space-time locality, so that synchronous migration of high-association objects is realized; a trend term, a season term and a residual term are extracted through time sequence analysis, a service load prediction function is constructed, a low load migration window is recognized, and priority scheduling is conducted on a high I / O task; model parameters are dynamically adjusted by monitoring performance jitter and response deviation degree, and hierarchical decision intelligence and system stability are improved.

Description

Technical Field

[0001] This invention relates to the field of computer data storage technology, specifically to a data intelligent hierarchical storage method and system based on multi-dimensional quantitative evaluation. Background Technology

[0002] With the deepening of enterprise digital transformation and the widespread adoption of IoT devices, the total amount of global data is experiencing explosive growth. The data generated in daily enterprise operations is not only massive in quantity but also exhibits characteristics of multi-source heterogeneity and dynamically changing access patterns. Data has become a core asset for enterprises, and the efficiency of its storage management directly impacts business response speed, operating costs, and market competitiveness. Against this backdrop, how to build an intelligent data management system that can meet high-performance access needs while effectively controlling storage costs has become a key technological challenge faced by enterprises in their digital transformation process.

[0003] Current data tiered storage methods suffer from several substantial drawbacks: First, static tiering strategies based on fixed rules fail to accurately reflect real-time changes in data value, often reserving frequently accessed critical business data in underperforming storage tiers or allowing rarely accessed historical data to occupy high-performance storage resources for extended periods. Second, existing methods typically treat individual data objects in isolation, neglecting the inherent relationships between data in business operations. This results in closely related data being scattered across different performance tiers, increasing access latency for business operations. Furthermore, traditional data migration timing lacks intelligent predictive capabilities, often triggering large-scale data migrations during peak business periods, exacerbating system resource contention and impacting normal business operations. These shortcomings make it difficult for existing storage systems to achieve the optimal balance between performance and cost in dynamically changing business environments. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a data intelligent hierarchical storage method and system based on multi-dimensional quantitative evaluation, thus solving the problems mentioned above.

[0005] To achieve the above objectives, this invention provides the following technical solution: a data intelligent hierarchical storage method based on multi-dimensional quantitative evaluation, comprising the following steps: S1. Real-time collection of dynamic access characteristics and static attribute characteristics of data objects, adaptive adjustment of feature weights based on the fluctuation intensity of data activity, and calculation of hierarchical migration coefficients of data objects; S2. Using the hierarchical migration coefficients as input, constructing a data association graph by analyzing co-occurrence relationships in business operation sequences, identifying sets of data objects with strong business associations through density-based clustering algorithms, establishing collaborative migration constraint rules based on the spatiotemporal locality characteristics of data access, and synchronously adjusting the set according to the constraint rules when the hierarchical migration coefficient of any object in the association set triggers a change in storage hierarchy. S3. Input the hierarchical migration coefficient and collaborative migration constraint rules into the migration decision unit, perform periodic decomposition of historical access data through time series analysis algorithm, extract trend items, seasonal items and residual items, construct business load prediction function, identify the period when the system resource utilization is lower than the set threshold as the migration window, decompose the high I / O load migration task into multiple sub-tasks and schedule execution within the identified migration window; S4. By monitoring the system performance jitter amplitude and business response time deviation during the migration process, when the root mean square value of the performance jitter amplitude exceeds the first threshold or the business response time deviation exceeds the second threshold, reduce the weight ratio of data activity and adjust the weight of the seasonal item in the business load prediction function.

[0006] Further, step S1 specifically includes the following steps: real-time acquisition of the access time series of data objects, extraction of access frequency, access time interval distribution and concurrent access volume as dynamic access features, and acquisition of the storage capacity, data structure complexity and predefined business criticality level of data objects as static attribute features; calculation of the coefficient of variation of data activity, and dynamic adjustment of the weight allocation ratio of dynamic access features and static attribute features based on the magnitude of the coefficient of variation; and calculation of the hierarchical migration coefficient of data objects through linear combination based on the weighted feature values, wherein the weight of dynamic access features increases with the increase of the coefficient of variation.

[0007] Furthermore, using the hierarchical migration coefficient as input, the logical process of constructing a data association graph by analyzing the co-occurrence relationships in the business operation sequence is as follows: Parse the business operation serial number, extract the data access sequence in the business operation, and calculate the co-occurrence frequency and temporal relationship of data objects in the business operation; construct an association matrix of data objects based on the co-occurrence frequency, where the matrix element values ​​represent the association strength between data objects; construct a weighted data association graph based on the association matrix, where nodes represent data objects and edge weights represent association strength; map the data association graph to a low-dimensional vector space using a graph embedding algorithm, preserving the association relationships between data objects.

[0008] Furthermore, the specific process of identifying sets of data objects with strong business relationships using density-based clustering algorithms and establishing collaborative migration constraint rules based on the spatiotemporal locality characteristics of data access is as follows: The OPTICS clustering algorithm is applied to the low-dimensional vector space of the data association graph to identify densely connected sets of data objects. The silhouette coefficient of each cluster set is calculated, and sets of data objects with strong business relationships are selected. The spatiotemporal characteristics of the data object sets are analyzed, and access time correlation and storage space proximity are extracted. Collaborative migration constraint rules are established based on the spatiotemporal locality characteristics, including: migration time sequence constraint: associated data objects complete migration within the same time period; storage location constraint: associated data objects maintain spatial proximity in the target storage layer; access path constraint: associated data objects maintain the original optimized access path after migration.

[0009] Furthermore, when the hierarchical migration coefficient of any object in the associated set triggers a storage level change, the specific process of synchronously adjusting the storage strategy of related objects in the set according to the constraint rules is as follows: Monitor the real-time changes of the hierarchical migration coefficients of all data objects in the associated set. When the migration coefficient of any data object reaches the storage level change threshold, trigger a collaborative migration evaluation of the associated set. Based on the association strength weight in the collaborative migration constraint rules, recalculate the target storage level of each data object in the set. Specifically, for data objects that are directly and strongly associated with the triggering object, adjust their target storage level to the same level as the triggering object. For data objects that are indirectly associated with the triggering object, calculate their recommended storage level based on the association strength weight. Based on the recalculated target storage level, generate a collaborative migration sequence to ensure that the associated data objects complete the level switch according to the preset timing constraints. During the migration execution process, dynamically adjust the execution order and concurrency of the migration tasks according to the system resource utilization.

[0010] Furthermore, the specific process of constructing a business load prediction function by periodically decomposing historical access data using time series analysis algorithms to extract trend, seasonal, and residual terms is as follows: Data access records within a historical time period are obtained; access volume is statistically analyzed at a fixed time granularity to form a time series; the time series is tested for stationarity, and non-stationary components are eliminated through differencing; the trend term is extracted using the moving average method to reflect the long-term variation pattern of access volume; major seasonal cycles are identified through periodogram analysis; seasonal components are extracted using Fourier transform; the residual term is calculated and its autocorrelation is tested; an autoregressive model is established to describe random fluctuations; the trend, seasonal, and residual terms are superimposed and reconstructed to form the business load prediction function.

[0011] Furthermore, the process of identifying periods when system resource utilization is below a set threshold as migration windows, and decomposing high I / O load migration tasks into multiple sub-tasks and scheduling their execution within the identified migration windows is as follows: Based on the business load prediction function, the expected value of system resource utilization is calculated, and periods when resource utilization is consistently below the set threshold are identified. Available migration windows are determined by combining business operation characteristics. Based on the spatiotemporal constraints in the collaborative migration constraint rules, high I / O load migration tasks are logically grouped, and each migration group is decomposed into multiple sub-tasks according to data block size and I / O load balancing principles. A sub-task priority evaluation mechanism is established based on hierarchical migration coefficients and business criticality. Within the determined migration window, sub-tasks are scheduled and executed according to priority order.

[0012] Further, step S4 includes the following steps: continuously collecting system performance indicators during the migration process, including CPU utilization, memory usage, and disk I / O response time; calculating the root mean square value of performance jitter; and statistically analyzing the deviation of business response time from the benchmark value. A performance evaluation matrix is ​​established, and performance indicators are compared and analyzed with preset thresholds. When the performance jitter exceeds the first threshold, the proportion of data activity in the feature weights is reduced proportionally according to the degree of deviation, while the weight of business criticality is increased proportionally. When the deviation of business response time exceeds the second threshold, the amplitude parameter of the seasonal term in the business load prediction function is adjusted according to the degree of deviation to reduce the impact range of periodic fluctuations.

[0013] A data-intelligent hierarchical storage system based on multi-dimensional quantitative evaluation includes the following modules: a data monitoring and weighting module, used to collect dynamic access characteristics and static attribute characteristics of data objects in real time, adaptively adjust feature weights according to the fluctuation intensity of data activity, and calculate the hierarchical migration coefficient of data objects; an association analysis and constraint module, used to take the hierarchical migration coefficient as input, construct a data association graph by analyzing the co-occurrence relationship in the business operation sequence, identify sets of data objects with strong business associations through density-based clustering algorithms, establish collaborative migration constraint rules based on the spatiotemporal locality characteristics of data access, and when the hierarchical migration coefficient of any object in the association set is detected to trigger a storage level change, the storage strategy of related objects in the set is synchronously adjusted according to the constraint rules; and a migration... The migration decision and scheduling module is used to input hierarchical migration coefficients and collaborative migration constraint rules into the migration decision unit. It uses time series analysis algorithms to periodically decompose historical access data, extract trend items, seasonal items, and residual items, construct a business load prediction function, identify periods when system resource utilization is lower than a set threshold as migration windows, decompose high I / O load migration tasks into multiple sub-tasks, and schedule their execution within the identified migration windows. The performance monitoring and optimization module is used to monitor the system performance jitter and business response time deviation during the migration process. When the root mean square value of the performance jitter exceeds the first threshold or the business response time deviation exceeds the second threshold, it reduces the weight of data activity and adjusts the weight of the seasonal item in the business load prediction function.

[0014] The present invention has the following beneficial effects: (1) The data intelligent hierarchical storage method based on multi-dimensional quantitative evaluation collects the dynamic access characteristics and static attribute characteristics of data objects in real time, and adaptively adjusts the feature weights based on the fluctuation intensity of data activity to accurately depict the data hierarchical migration trend, overcoming the problem of slow response under sudden changes in access patterns in the traditional fixed threshold hierarchical method. Furthermore, by analyzing the co-occurrence relationship in the business operation sequence to construct a data association graph, and using a density-based clustering algorithm to identify sets of strongly business-related objects, it can maintain the logical consistency and access continuity between data during the hierarchical migration process, avoid access interruption and redundant I / O operations caused by single object migration, thereby improving the stability and business matching degree of the hierarchical storage strategy.

[0015] (2) The data intelligent tiered storage system based on multi-dimensional quantitative evaluation inputs the tiered migration coefficient and collaborative migration constraint rules into the migration decision unit. Through time series analysis algorithms, historical access data is periodically decomposed to establish a prediction function oriented towards changes in business load. This enables adaptive scheduling of migration tasks during periods of low system load, effectively reducing resource contention and migration conflicts under high concurrency conditions. Simultaneously, by monitoring the performance fluctuations and deviations in business response time during the migration process, the system dynamically adjusts the weights of data activity and seasonal items, forming a closed-loop feedback optimization mechanism. This ensures that the system can stably maintain performance balance and resource utilization efficiency during continuous operation, thereby significantly improving the adaptability and long-term operational reliability of the intelligent storage system.

[0016] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0017] Figure 1 This is a flowchart of the intelligent hierarchical data storage method based on multi-dimensional quantitative evaluation according to the present invention.

[0018] Figure 2 This is a flowchart of the intelligent hierarchical storage system for data based on multi-dimensional quantitative evaluation, as described in this invention. Detailed Implementation

[0019] This application's embodiments address the problems of existing layered storage methods failing to dynamically reflect the real-time value of data, neglecting the business correlation between data, and experiencing large system performance fluctuations during migration and scheduling by providing a data intelligent layered storage method and system based on multi-dimensional quantitative evaluation.

[0020] The overall concept of the solution in this application embodiment is as follows: First, a multi-dimensional quantitative evaluation model is used to acquire the dynamic access characteristics and static attribute characteristics of data objects in real time. Weights are adaptively adjusted based on the intensity of activity fluctuations, and the hierarchical migration coefficient of data objects is calculated, achieving dynamic perception and accurate evaluation of data levels. Second, a data association graph is constructed based on the co-occurrence relationships of business operation sequences. Density clustering algorithms are used to identify strongly correlated data sets, and collaborative migration constraint rules are established to maintain business consistency during the hierarchical adjustment process. Third, a time series analysis algorithm is used to decompose historical access data into trends, constructing a business load prediction function. Migration tasks are executed during periods of low system resource utilization, optimizing the timing and granularity of migration scheduling. Finally, a feedback loop is formed by monitoring performance fluctuations and response deviations during the migration process, dynamically correcting weight allocation and prediction model parameters, thereby achieving adaptive optimization of the data hierarchical strategy and long-term stable operation of the system performance.

[0021] Please see Figure 1This invention provides a technical solution: a data intelligent hierarchical storage method based on multi-dimensional quantitative evaluation, comprising the following steps: S1. Real-time collection of dynamic access characteristics and static attribute characteristics of data objects, adaptive adjustment of feature weights according to the fluctuation intensity of data activity, and calculation of hierarchical migration coefficients of data objects; S2. Using the hierarchical migration coefficients as input, constructing a data association graph by analyzing co-occurrence relationships in business operation sequences, identifying sets of data objects with strong business associations through density-based clustering algorithms, establishing collaborative migration constraint rules based on the spatiotemporal locality characteristics of data access, and synchronously adjusting related pairs within the set according to the constraint rules when the hierarchical migration coefficient of any object in the association set triggers a change in storage level. S3. Input the hierarchical migration coefficient and collaborative migration constraint rules into the migration decision unit, perform periodic decomposition of historical access data through time series analysis algorithm, extract trend item, seasonal item and residual item, construct business load prediction function, identify the period when the system resource utilization is lower than the set threshold as the migration window, decompose the high I / O load migration task into multiple sub-tasks and schedule execution within the identified migration window; S4. By monitoring the system performance jitter amplitude and business response time deviation during the migration process, when the root mean square value of the performance jitter amplitude exceeds the first threshold or the business response time deviation exceeds the second threshold, reduce the weight ratio of data activity and adjust the weight of the seasonal item in the business load prediction function.

[0022] In this implementation scheme, S1. In this step, dynamic access characteristics refer to behavioral indicators such as the access frequency, access duration, and read / write ratio of data during system operation, which change over time; static attribute characteristics refer to relatively fixed information such as data type, business affiliation, and security level. The system collects the above multi-dimensional characteristics and uses a weighted evaluation model to calculate the activity change trend of data objects in different time periods. The activity fluctuation intensity is used to measure the stability of data access behavior and is obtained by statistically analyzing the time variation variance of access volume. Based on the magnitude of the fluctuation intensity, the system adaptively adjusts the weight ratio of each feature to generate a hierarchical migration coefficient that reflects the real-time value of the data. This coefficient is used to determine the migration priority of data between different storage levels. S2. In this step, co-occurrence relationship refers to the statistical regularity of different data objects being accessed simultaneously or sequentially in the business operation sequence, which is used to reflect the degree of logical coupling between data. The system constructs a data association graph based on the co-occurrence frequency. This graph uses nodes to represent data objects, edges to represent access associations, and edge weights to represent co-occurrence intensity. Using a density-based clustering algorithm to cluster the graph structure, data sets with similar access characteristics and high business relevance can be automatically identified. Furthermore, the system establishes collaborative migration constraint rules based on the spatiotemporal locality of data access (i.e., the concentration of access in time and space dimensions). This ensures that when any highly active data in the set triggers a hierarchical change, its associated data can be migrated synchronously, preventing performance loss and I / O jitter caused by cross-level access. S3. In this step, the migration decision unit is the core control module of the system, used to formulate a migration plan based on historical access patterns and real-time load prediction results. The time series analysis algorithm can use an additive decomposition model to break down historical access data into long-term trend changes (trend term), periodic fluctuations (seasonal term), and random disturbances (residual term), thereby forming a predictable business load curve. Based on this curve, the system identifies idle time periods as migration windows and performs subtask-level decomposition and distributed scheduling of high I / O load tasks to reduce the impact on the main business flow during migration. This design enables smooth transition of storage resources and optimized migration energy consumption. S4. In this step, performance jitter refers to the intensity of fluctuations in system I / O throughput or CPU utilization during the migration process, obtained by calculating the root mean square value of performance indicator changes over a period of time; business response time deviation refers to the degree of deviation of the system response time from the steady state. The system monitors these two indicators in real time. When a threshold is exceeded, an adaptive correction mechanism is automatically triggered to reduce the influence weight of activity characteristics, thereby suppressing frequent migrations caused by misjudged active data; at the same time, the weight of the seasonal term in the prediction model is adjusted to make the system more robust to periodic business load fluctuations, achieving closed-loop optimization and dynamic stability control of the layered strategy.In summary, this embodiment achieves dynamic optimization of data tiered storage and long-term balance of system performance through a collaborative mechanism of multi-dimensional quantitative evaluation, business correlation analysis, time series prediction, and feedback adaptive adjustment, providing reliable technical support for intelligent data management in complex business environments.

[0023] Specifically, step S1 includes the following steps: real-time acquisition of the access time series of data objects, extraction of access frequency, access time interval distribution and concurrent access volume as dynamic access features, and acquisition of the storage capacity, data structure complexity and predefined business criticality level of data objects as static attribute features; calculation of the coefficient of variation of data activity, and dynamic adjustment of the weight allocation ratio of dynamic access features and static attribute features based on the magnitude of the coefficient of variation; and calculation of the hierarchical migration coefficient of data objects through linear combination based on the weighted feature values, wherein the weight of dynamic access features increases with the increase of the coefficient of variation.

[0024] In this implementation plan, the system first collects the access time series of data objects in real time, extracting access frequency, access time interval distribution, and concurrent access volume as dynamic access features; simultaneously, it acquires the storage capacity, data structure complexity, and predefined business criticality level of the data objects as static attribute features. Dynamic access features reflect the activity level of data over time, while static attribute features describe the importance and resource consumption of data in the business dimension. Subsequently, the coefficient of variation of data activity is calculated, i.e., the ratio of the standard deviation of access frequency to its mean, to characterize the volatility of access behavior. The calculation formula is as follows: ;in, The activity variation coefficient of a data object is used to reflect the intensity of fluctuations in access frequency. The standard deviation of access frequency indicates the degree of dispersion of access behavior; The mean access frequency represents the overall access level. Based on the coefficient of variation, the system dynamically adjusts the weighting ratio between dynamic access features and static attribute features. The weighting adjustment function can adopt an exponential smoothing form. ; ;in, Weights of dynamic access features; Weights of static attribute features; The weighting adjustment coefficient controls the sensitivity of the coefficient of variation to weight allocation. It can be determined offline by fitting the value where the system performance fluctuation is minimized in historical migration tasks. After determining the weights, the hierarchical migration coefficient of the data objects is calculated through a linear combination based on the weighted feature values. This coefficient represents the data migration priority and hierarchical tendency. The calculation formula is: ;in, : Hierarchical migration coefficient of data objects; :No. The dynamic access characteristics include access frequency, time interval distribution characteristics, and concurrent access volume; :No. The static attribute characteristics of the item include storage capacity, structural complexity, and business criticality level; The normalized weight coefficients of dynamic access features are determined by principal component analysis based on the feature variance contribution rate. The normalized weight coefficients of static attribute features are determined based on the indicator weights in the business importance assessment model; : These represent the number of dynamic access features and static attribute features, respectively. During the above calculation process, the system continuously updates the access time series and... A sliding window calculation is used to enable real-time weight adjustment and hierarchical migration coefficient evaluation. When When the weights increase, dynamically access feature weights The corresponding improvements enable the system to pay more attention to data with frequently changing access patterns; when When the data is small and access behavior is stable, the influence of static attribute features is enhanced, thereby achieving an adaptive balance in data hierarchy evaluation.

[0025] Specifically, the logical process of constructing a data association graph by analyzing the co-occurrence relationships in the business operation sequence, using the hierarchical migration coefficient as input, is as follows: Parse the business operation serial number, extract the data access sequence in the business operation, and calculate the co-occurrence frequency and temporal relationship of data objects in the business operation; construct an association matrix of data objects based on the co-occurrence frequency, where the matrix element values ​​represent the association strength between data objects; construct a weighted data association graph based on the association matrix, where nodes represent data objects and edge weights represent association strength; and map the data association graph to a low-dimensional vector space using a graph embedding algorithm, preserving the association relationships between data objects.

[0026] In this implementation scheme, a graph embedding algorithm based on random walks is used to achieve low-dimensional representation of data objects. Specifically, for any node in the data association graph... Its context node The co-occurrence probability is determined by the normalized transition matrix, and the embedding objective function is defined as: ;in, For nodes With nodes Edge weights between them; Let be the co-occurrence probability of two nodes in the embedding space, given by Sure; These represent the embedding vectors of the nodes. The objective function is minimized. This yields the optimal representation of each node in the low-dimensional space. To prevent embedding distortion caused by extreme imbalance of weight coefficients, [the following is done]... Introducing a smoothing factor ,Right now ,in These are the original edge weights. An empirical value of 0.5 to 0.8 is used, and the optimal value is selected through cross-validation. This method can enhance the expressive power of weakly connected nodes while maintaining the global structure of the graph.

[0027] Specifically, the process of identifying sets of data objects with strong business relationships using density-based clustering algorithms and establishing collaborative migration constraint rules based on the spatiotemporal locality characteristics of data access is as follows: The OPTICS clustering algorithm is applied to the low-dimensional vector space of the data association graph to identify densely connected sets of data objects. The silhouette coefficient of each cluster is calculated to select sets of data objects with strong business relationships. The spatiotemporal characteristics of the data object sets are analyzed, and access time correlation and storage space proximity are extracted. Collaborative migration constraint rules are established based on the spatiotemporal locality characteristics, including: migration time sequence constraint: associated data objects complete migration within the same time period; storage location constraint: associated data objects maintain spatial proximity in the target storage layer; access path constraint: associated data objects maintain the original optimized access path after migration.

[0028] In this implementation scheme, collaborative migration constraint rules are constructed based on density characteristics and spatiotemporal locality. Specifically, for any data object in the low-dimensional space... Its local density and reachable distance The calculation is as follows: , ;in, For object and Euclidean distance; This is the kernel function bandwidth parameter. According to... and The distribution determines the set of objects with peak density, which is used to form the initial cluster core. Based on this, the temporal correlation parameters of the objects are calculated according to the access records. and spatial proximity parameters Construct comprehensive locality indicators: ;in, and These are the time and space weighting coefficients, determined by minimizing the average distance within each cluster and the cross-cluster variance ratio. To establish collaborative migration constraints: when objects within a cluster set... Similarity higher than the threshold At that time, a unified migration timing and spatial location allocation are performed. Threshold According to the sliding window The mean jz and standard deviation of the distribution Determined, the calculation formula is: ,in An adjustment coefficient is used to control migration sensitivity. Next, based on the redefined target storage tier, the system generates a collaborative migration sequence. The migration sequence is constructed according to the following constraints: Timing constraints: ensuring that data objects directly dependent on business operations complete migration within the same time window to avoid access link breaks; Location constraints: prioritizing the allocation of data objects with high correlation strength to physically adjacent storage nodes to reduce cross-tier access latency; Resource constraints: based on the system's real-time resource utilization metrics. (Including CPU utilization, I / O queue length, etc.), dynamically adjust the concurrency of migration tasks. Its update rules are as follows: ;in, To adjust the sensitivity parameters; The reference resource utilization threshold is used to maintain system operational stability. Through the above calculation and constraint mechanisms, this implementation scheme can achieve synchronous migration decisions and adaptive resource scheduling for the entire associated set when any data object hierarchical change trigger condition is detected, thereby balancing system performance stability and dynamic balance of storage tiers.

[0029] Specifically, when the hierarchical migration coefficient of any object in the associated set triggers a storage level change, the process of synchronously adjusting the storage strategy of related objects in the set according to the constraint rules is as follows: Monitor the real-time changes in the hierarchical migration coefficients of all data objects in the associated set. When the migration coefficient of any data object reaches the storage level change threshold, trigger a collaborative migration assessment of the associated set. Based on the association strength weight in the collaborative migration constraint rules, recalculate the target storage level of each data object in the set. For data objects with a direct and strong association with the triggering object, adjust their target storage level to the same level as the triggering object. For data objects with an indirect association with the triggering object, calculate their recommended storage level based on the association strength weight. Based on the recalculated target storage level, generate a collaborative migration sequence to ensure that associated data objects complete the level switch according to the preset timing constraints. During the migration execution process, dynamically adjust the execution order and concurrency of migration tasks according to system resource utilization.

[0030] In this implementation scheme, firstly, the target storage level for each data object within the set is recalculated based on the association strength weight. The calculation formula is as follows: ;in, Represents data objects Target level value; For object With object The correlation strength weight between them; Representation Object The current layer level encoding value (e.g., different encoding values ​​for high-speed layer, medium-speed layer, and low-speed layer); the summation range is the sum of the values ​​of the objects. All objects that are directly or indirectly related. Weight coefficient. The method for determining it is as follows: ;in, For object With object The edge weight represents the strength of the association between the two in business operations; For object The total association strength with all associated objects is used for normalization to prevent strong associations with a particular object from biasing the results. The difference between the value and the current level's encoding exceeds the level change threshold. When this happens, the system determines that the object should be migrated to a new level. Threshold Determined through dynamic historical statistical methods, the calculation formula is: Where qw is the average value of the object hierarchy over the most recent period; Its standard deviation; The adjustment coefficient is obtained by minimizing the ratio of migration frequency to performance gain. Next, based on the redefined target storage tier, the system generates a cooperative migration sequence. This migration sequence is constructed according to cooperative migration constraint rules.

[0031] Specifically, the process of constructing a business load prediction function by periodically decomposing historical access data using time series analysis algorithms to extract trend, seasonal, and residual terms is as follows: Data access records within a historical time period are obtained; access volume is statistically analyzed at a fixed time granularity to form a time series; the time series is tested for stationarity, and non-stationary components are eliminated through differencing; the trend term is extracted using the moving average method to reflect the long-term variation pattern of access volume; major seasonal cycles are identified through periodogram analysis; seasonal components are extracted using Fourier transform; the residual term is calculated and its autocorrelation is tested; an autoregressive model is established to describe random fluctuations; the trend, seasonal, and residual terms are superimposed and reconstructed to form the business load prediction function.

[0032] In this implementation plan, firstly, data access records within a historical time period are obtained, and then a time series is formed according to a fixed time granularity. The time series was subjected to a stationarity test. For non-stationary series, a differencing operator was used to eliminate non-stationary components until the series met the stationarity requirement. Subsequently, the time series was smoothed using the moving average method to obtain a trend term reflecting the long-term trend. The period length of significant periods is identified through periodogram and spectral analysis. Then, the periodic fluctuation component is extracted as the seasonal term using a finite-term Fourier series approximation. Finally, the residual term is obtained by subtracting the difference between the trend term and the seasonal term from the original sequence. An autoregressive model is used to model its random fluctuations. By superimposing the three parts, a complete business load prediction function is obtained, which is used to predict traffic at future times. Its mathematical expression is: ;in, For a moment The number of visits, For trend items, For seasonal items, This is the residual term. The trend term is calculated using the moving average operator: ;in, Indicates the window length is Simple moving average operator, window length The seasonal term is determined by calculating the variance of the smoothed residuals under different candidate lengths and selecting the minimum value. The seasonal term is expressed in a finite-term Fourier expansion form. ;in, To identify the main cycles through periodogram analysis. The number of terms in the Fourier expand is determined by selecting a few frequencies whose spectral energy percentage reaches a set threshold, and the coefficients are... and Obtained by fitting the historical sequence using the least squares method. The residual term is described using an autoregressive model: ;in, The order of autoregression is selected using the AIC or BIC criteria. These are the autoregressive coefficients. This represents a random disturbance term that follows a white noise distribution. Window length. Fourier terms and autoregressive order The values ​​are all determined with the goal of minimizing prediction error, through cross-validation or information criteria; the residual model coefficients are updated using recursive least squares to ensure that the model can adapt to business fluctuations. The resulting predictions are then used to generate a system resource utilization expectation sequence through sliding window statistics, providing input for the IP of the migration window.

[0033] Specifically, the process of identifying periods when system resource utilization is below a set threshold as migration windows, and decomposing high I / O load migration tasks into multiple sub-tasks and scheduling their execution within the identified migration windows is as follows: Based on the business load prediction function, the expected value of system resource utilization is calculated, and periods when resource utilization is consistently below the set threshold are identified. Available migration windows are determined by combining business operation characteristics. Based on the spatiotemporal constraints in the collaborative migration constraint rules, high I / O load migration tasks are logically grouped, and each migration group is decomposed into multiple sub-tasks according to data block size and I / O load balancing principles. A sub-task priority evaluation mechanism is established based on hierarchical migration coefficients and business criticality. Within the determined migration windows, sub-tasks are scheduled and executed according to priority.

[0034] In this implementation plan, the expected value sequence of system resource utilization is calculated based on the aforementioned business load prediction function. The system identifies consecutive periods of low load within this sequence as candidate migration windows. A migration window threshold is defined by statistically analyzing the utilization distribution during the forecast period. ;in, This represents the average resource utilization rate over the forecast period. Standard deviation, This is the migration sensitivity coefficient, used to control the magnitude of threshold reduction; this coefficient is determined by an objective function that minimizes the trade-off between business impact and migration efficiency on historical migration samples. It is applied when the length of the continuous time period is not less than the minimum migration window length. And satisfy When that time period is determined to be an available migration window, then... The migration window is estimated based on the average time taken for a single migration and the cost of concurrent system switching. After determining the migration window, the high I / O data to be migrated is logically grouped according to the collaborative migration set, and then further divided into several subtasks based on data block size and I / O load balancing principles. Each subtask is assigned a priority score. Calculated by comprehensively considering hierarchical migration coefficients, business criticality, and data block size: ;in, The average hierarchical migration coefficient of the set corresponding to the subtasks. The average business criticality score is given. For data block size, The priority weight coefficient, and satisfies The initial weight values ​​are obtained through historical migration effect optimization and are fine-tuned periodically during operation to match the real-time strategy. The system adjusts the weights within the migration window according to... Subtasks are scheduled for execution in descending order of resource utilization, and the concurrency is dynamically adjusted based on real-time resource utilization metrics. ;in, As the baseline concurrency, As a reference threshold for resource utilization, To adjust the sensitivity coefficient concurrently; The value of is determined by analyzing the balance between business response fluctuations and migration rate during the migration process. Through the above prediction, identification, and scheduling mechanisms, adaptive migration window planning based on load prediction and hierarchical scheduling of high I / O tasks are realized, thereby improving the efficiency of hierarchical data migration and system resource utilization while ensuring business stability.

[0035] Specifically, step S4 includes the following steps: continuously collecting system performance metrics during the migration process, including CPU utilization, memory usage, and disk I / O response time; calculating the root mean square value of performance jitter; and statistically analyzing the deviation of business response time from the baseline value. A performance evaluation matrix is ​​established, and performance metrics are compared with preset thresholds. When the performance jitter exceeds the first threshold, the proportion of data activity in the feature weights is reduced proportionally according to the degree of deviation, while the weight of business criticality is increased proportionally. When the deviation of business response time exceeds the second threshold, the amplitude parameter of the seasonal term in the business load prediction function is adjusted according to the degree of deviation to reduce the impact range of periodic fluctuations.

[0036] In this implementation scheme, the threshold determination and adaptive triggering rule are as follows: Let the first threshold be... (against The second threshold is (against The adaptive determination method for the two thresholds is as follows: Historical performance and response data are collected during the system baseline period, the mean and standard deviation of the corresponding statistics are calculated, and then a robust threshold strategy is adopted: , ;in, During the baseline period The mean and standard deviation; During the baseline period The mean and standard deviation; Threshold amplification factor, used to adjust trigger sensitivity; its value is determined through offline simulation with the goal of minimizing business impact, and can be subject to upper / lower limits at runtime based on the system's SLA strategy. When detected... At that time, the feature weights are adaptively adjusted according to the degree of deviation. Let the original dynamic access feature weights be... Business criticality weight is The adjustment rule is to reduce / increase proportionally: ;in, Root mean square deviation ratio; : Weight adjustment sensitivity coefficient; its value is selected by grid search or Bayesian optimization on historical migration events to minimize system recovery time and migration rollbacks after triggering. Weight value before triggering; The adjusted weight values. The above expression ensures weight conservation (i.e., the total weight is not lost unnecessarily with a single adjustment) and migrates weights from activity features to business criticality features to reduce misjudgments of business sensitivity due to migration frequency. Initial weight values ​​and adjustment limits are determined through simulation, and over-adjustment is avoided during system operation through rule-based restrictions. When a... At that time, the amplitude parameter of the seasonal term in the business load forecast function is adjusted according to the degree of deviation to reduce the impact of periodic components on the forecast. Let the original seasonal amplitude be... The adjustment rules are as follows: ;in, : Response deviation ratio; v: Amplitude adjustment sensitivity coefficient; its value is determined by offline measurement while taking into account both prediction robustness and migration accuracy; : Amplitude parameters before and after seasonal adjustment. To avoid excessive model oscillation caused by a single anomaly, a lag and smoothing strategy is introduced: the actual effective values ​​of weights and amplitudes are updated using exponential smoothing, i.e., the new effective values ​​are the weighted average of historical effective values ​​and candidate adjustment values. The smoothing coefficient is configured by the system administrator or adaptively determined by minimizing migration jitter.

[0037] Please see Figure 2The data intelligent hierarchical storage system based on multi-dimensional quantitative evaluation includes the following modules: a data monitoring and weighting module, which collects the dynamic access characteristics and static attribute characteristics of data objects in real time, adaptively adjusts the feature weights according to the fluctuation intensity of data activity, and calculates the hierarchical migration coefficient of data objects; and an association analysis and constraint module, which takes the hierarchical migration coefficient as input, constructs a data association graph by analyzing the co-occurrence relationship in the business operation sequence, identifies a set of data objects with strong business associations through a density-based clustering algorithm, establishes collaborative migration constraint rules based on the spatiotemporal locality characteristics of data access, and synchronously adjusts the storage strategy of related objects in the set according to the constraint rules when the hierarchical migration coefficient of any object in the association set triggers a change in storage level. The migration decision scheduling module is used to input hierarchical migration coefficients and collaborative migration constraint rules into the migration decision unit. It uses time series analysis algorithms to periodically decompose historical access data, extract trend items, seasonal items, and residual items, construct a business load prediction function, identify periods when system resource utilization is lower than a set threshold as migration windows, decompose high I / O load migration tasks into multiple sub-tasks, and schedule their execution within the identified migration windows. The performance monitoring and optimization module is used to monitor the system performance jitter and business response time deviation during the migration process. When the root mean square value of the performance jitter exceeds the first threshold or the business response time deviation exceeds the second threshold, it reduces the weight of data activity and adjusts the weight of the seasonal item in the business load prediction function.

[0038] In this implementation scheme, the data monitoring weight module is used to achieve comprehensive quantification and dynamic weight adjustment of multi-dimensional features. By real-time collection of dynamic features such as data object access frequency, access interval, and concurrent access volume, and synchronous monitoring of static features such as storage capacity, structural complexity, and business criticality level, the system calculates the fluctuation intensity of data activity. When the activity variation coefficient increases, the system automatically increases the weight ratio of dynamic features, thereby highlighting the changing trend of data access behavior when calculating the hierarchical migration coefficient, achieving sensitive capture and real-time updates of data hot and cold changes. The association analysis constraint module is used to establish business dependencies between data before hierarchical migration. By analyzing the co-occurrence patterns of business operation sequences, a data association graph is constructed, and a set of data objects with high correlation is identified based on a density clustering algorithm. Subsequently, based on the spatiotemporal locality characteristics of data access, collaborative migration constraint rules are defined, including strategies such as synchronous migration of strongly correlated objects and hierarchical migration of indirectly correlated objects. When the hierarchical migration coefficient of any object in the set triggers a hierarchical change, the module automatically adjusts the storage level of all related data in the set according to the constraint rules, ensuring access consistency and migration synergy. The migration decision-making and scheduling module is used to predict system resource status and optimize tasks before executing migration operations. The module inputs hierarchical migration coefficients and collaborative constraint rules into the decision unit, uses a time series decomposition algorithm to extract trend, seasonal, and residual terms of access volume, constructs a business load prediction function, and predicts changes in system resource utilization. Based on the prediction results, it identifies time periods where resource utilization is below a set threshold as migration windows, and decomposes the migration task into multiple sub-tasks according to task I / O load intensity. These sub-tasks are then scheduled and executed in parallel according to priority within the migration window, thereby improving migration efficiency while ensuring performance stability. The performance monitoring and optimization module is used to dynamically perceive the system's operating status and provide feedback on optimization strategies during migration execution. The module continuously monitors performance indicators such as CPU utilization, memory usage, and disk I / O response time, and calculates the root mean square value of performance jitter and the deviation from the business response time. When performance fluctuations exceed the first threshold, the system reduces the feature weight of data activity proportional to the degree of deviation and correspondingly increases the weight of business criticality to suppress performance fluctuations caused by migration. When the deviation of business response time exceeds the second threshold, the system automatically adjusts the amplitude parameter of the seasonal term in the business load prediction function to reduce the impact of periodic fluctuations. Through this dynamic adjustment mechanism, real-time self-optimization of the migration strategy and adaptive control of system performance are achieved, making the entire tiered storage process more stable and intelligent.

[0039] In summary, this application has at least the following effects: A data-intelligent tiered storage method and system based on multi-dimensional quantitative evaluation achieves intelligent migration and dynamic optimization of data objects across different storage tiers by introducing a comprehensive multi-dimensional evaluation of data activity, business criticality, and access spatiotemporal characteristics. This solution significantly improves storage resource utilization and system response efficiency while ensuring business continuity. Through a collaborative migration constraint mechanism and load prediction scheduling strategy, it avoids resource contention and performance fluctuations in high-concurrency scenarios, enhancing the overall stability and adaptability of the system. Furthermore, by adjusting weight allocation based on performance monitoring feedback, it achieves closed-loop optimization of the migration strategy, enabling the tiered storage process to possess continuous self-learning and dynamic adjustment capabilities.

[0040] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0041] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0042] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0043] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0044] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0045] 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 the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A data intelligent hierarchical storage method based on multi-dimensional quantitative evaluation, characterized in that, Includes the following steps: S1. Real-time collection of dynamic access characteristics and static attribute characteristics of data objects, adaptive adjustment of feature weights based on the fluctuation intensity of data activity, and calculation of hierarchical migration coefficients of data objects; S2. Using the hierarchical migration coefficient as input, a data association graph is constructed by analyzing the co-occurrence relationship in the business operation sequence. A density-based clustering algorithm is used to identify a set of data objects with strong business associations. Based on the spatiotemporal locality characteristics of data access, collaborative migration constraint rules are established. When the hierarchical migration coefficient of any object in the association set is detected to trigger a change in storage level, the storage strategy of related objects in the set is adjusted synchronously according to the constraint rules. S3. Input the hierarchical migration coefficient and collaborative migration constraint rules into the migration decision unit, perform periodic decomposition on the historical access data through time series analysis algorithm, extract trend items, seasonal items and residual items, construct a business load prediction function, identify the period when the system resource utilization is lower than the set threshold as the migration window, decompose the high I / O load migration task into multiple sub-tasks and schedule the execution within the identified migration window. S4. By monitoring the system performance jitter and business response time deviation during the migration process, when the root mean square value of the performance jitter exceeds the first threshold or the business response time deviation exceeds the second threshold, the weight of data activity is reduced and the seasonal term weight of the business load prediction function is adjusted.

2. The intelligent hierarchical data storage method based on multi-dimensional quantitative evaluation according to claim 1, characterized in that: Step S1 specifically includes the following steps: Real-time acquisition of access time series of data objects, extraction of access frequency, access time interval distribution and concurrent access volume as dynamic access features, and acquisition of storage capacity, data structure complexity and predefined business criticality level of data objects as static attribute features; Calculate the coefficient of variation of data activity, and dynamically adjust the weighting ratio of dynamic access features and static attribute features based on the magnitude of the coefficient of variation. Based on the weighted feature values, the hierarchical migration coefficient of the data object is calculated through linear combination, where the weight of the dynamic access feature increases with the increase of the coefficient of variation.

3. The intelligent hierarchical data storage method based on multi-dimensional quantitative evaluation according to claim 1, characterized in that: The logical process of constructing a data association graph by using hierarchical migration coefficients as input and analyzing co-occurrence relationships in business operation sequences is as follows: Parse the business operation serial number, extract the data access sequence in the business operation, and calculate the co-occurrence frequency and temporal relationship of data objects in the business operation; An association matrix of data objects is constructed based on co-occurrence frequency, where the matrix element values ​​represent the association strength between data objects. A weighted data association graph is constructed based on the association matrix, where nodes represent data objects and edge weights represent association strength. The graph embedding algorithm maps the data association graph to a low-dimensional vector space, preserving the relationships between data objects.

4. The intelligent hierarchical data storage method based on multi-dimensional quantitative evaluation according to claim 3, characterized in that: The specific process of identifying sets of data objects with strong business relationships using density-based clustering algorithms and establishing collaborative migration constraint rules based on the spatiotemporal locality characteristics of data access is as follows: The OPTICS clustering algorithm is applied to the low-dimensional vector space of the data association graph to identify densely connected sets of data objects, calculate the silhouette coefficient of each cluster set, and filter out sets of data objects with strong business associations. Analyze the spatiotemporal characteristics of data object sets, extract access time correlation and storage space proximity, and establish collaborative migration constraint rules based on spatiotemporal locality characteristics, including: Migration timing constraints: Related data objects must complete the migration within the same time period; Storage location constraints: Associated data objects maintain spatial proximity in the target storage layer; Access path constraints: The access path of associated data objects is optimized to maintain the original access path after migration.

5. The intelligent hierarchical data storage method based on multi-dimensional quantitative evaluation according to claim 4, characterized in that: When a storage hierarchy change is triggered by the hierarchical migration coefficient of any object in the associated set, the specific process of synchronously adjusting the storage strategy of related objects in the set according to the constraint rules is as follows: Monitor the real-time changes in the hierarchical migration coefficients of all data objects in the associated set. When the migration coefficient of any data object reaches the storage level change threshold, trigger a collaborative migration assessment of the associated set. Based on the association strength weight in the collaborative migration constraint rules, the target storage level of each data object in the set is recalculated. Specifically: for data objects that are directly and strongly associated with the triggering object, their target storage level is adjusted to the same level as the triggering object; for data objects that are indirectly associated with the triggering object, their recommended storage level is calculated based on the association strength weight. Based on the recalculated target storage level, a collaborative migration sequence is generated to ensure that associated data objects complete the level switch according to the preset timing constraints. During the migration process, the execution order and concurrency of migration tasks are dynamically adjusted based on system resource utilization.

6. The intelligent hierarchical data storage method based on multi-dimensional quantitative evaluation according to claim 1, characterized in that: The specific process of using time series analysis algorithms to periodically decompose historical access data, extract trend items, seasonal items, and residual items, and construct a business load prediction function is as follows: Obtain data access records within a historical time period, statistically analyze access volume at a fixed time granularity to form a time series, perform a stationarity test on the time series, and eliminate non-stationary components through differencing operations; The trend term is extracted by using the moving average method to reflect the long-term change pattern of visit volume, the main seasonal cycle is identified by periodic chart analysis, and the seasonal component is extracted by Fourier transform. The residual term is calculated and its autocorrelation is tested. An autoregressive model is established to describe random fluctuations. The trend term, seasonal term and residual term are superimposed and reconstructed to form the business load forecasting function.

7. The intelligent hierarchical data storage method based on multi-dimensional quantitative evaluation according to claim 1, characterized in that: The process of identifying periods when system resource utilization is below a set threshold as migration windows, and decomposing high I / O load migration tasks into multiple subtasks and scheduling their execution within the identified migration windows is as follows: Based on the business load prediction function, the expected value of system resource utilization is calculated, the period when the resource utilization is continuously lower than the set threshold is identified, and the available migration window is determined in combination with the business operation characteristics. Based on the spatiotemporal constraints in the collaborative migration constraint rules, high I / O load migration tasks are logically grouped, and each migration group is decomposed into multiple sub-tasks according to the data block size and I / O load balancing principle. A subtask priority evaluation mechanism is established based on hierarchical migration coefficients and business criticality. Within a defined migration window, subtasks are scheduled and executed in order of priority.

8. The intelligent hierarchical data storage method based on multi-dimensional quantitative evaluation according to claim 1, characterized in that: Step S4 includes the following steps: During the migration process, system performance metrics are continuously collected, including CPU utilization, memory usage, and disk I / O response time. The root mean square value of performance jitter is calculated, and the degree of deviation of business response time from the baseline value is statistically analyzed. Establish a performance evaluation matrix, compare and analyze performance indicators with preset thresholds. When the performance fluctuation exceeds the first threshold, reduce the proportion of data activity in the feature weight according to the degree of deviation, and increase the weight of business criticality according to the corresponding proportion. When the deviation of the business response time exceeds the second threshold, the amplitude parameter of the seasonal term in the business load forecast function is adjusted according to the degree of deviation, thereby reducing the impact range of periodic fluctuations.

9. A data intelligent hierarchical storage system based on multi-dimensional quantitative evaluation, applied to the data intelligent hierarchical storage method based on multi-dimensional quantitative evaluation as described in any one of claims 1-8, characterized in that, Includes the following modules: The data monitoring weight module is used to collect dynamic access characteristics and static attribute characteristics of data objects in real time, adaptively adjust the feature weights according to the fluctuation intensity of data activity, and calculate the hierarchical migration coefficient of data objects. The association analysis constraint module is used to take the hierarchical migration coefficient as input, construct a data association graph by analyzing the co-occurrence relationship in the business operation sequence, identify a set of data objects with strong business associations through a density-based clustering algorithm, establish collaborative migration constraint rules based on the spatiotemporal locality characteristics of data access, and when the hierarchical migration coefficient of any object in the association set is detected to trigger a change in storage level, the storage strategy of related objects in the set is adjusted synchronously according to the constraint rules. The migration decision and scheduling module is used to input the hierarchical migration coefficients and collaborative migration constraint rules into the migration decision unit, perform periodic decomposition of historical access data through time series analysis algorithms, extract trend items, seasonal items and residual items, construct a business load prediction function, identify the period when the system resource utilization is lower than the set threshold as the migration window, decompose the high I / O load migration task into multiple sub-tasks and schedule the execution within the identified migration window. The performance monitoring and optimization module is used to monitor the system performance jitter and business response time deviation during the migration process. When the root mean square value of the performance jitter exceeds the first threshold or the business response time deviation exceeds the second threshold, the weight of data activity is reduced and the seasonal term weight of the business load prediction function is adjusted.

Images