Data block relevance data storage method for improving network security availability

Abstract

The invention discloses a data block relevance data storage method for improving network security availability, which comprises the following steps of: firstly, analyzing a historical IO (Input / Output) request, discovering a data block set which is frequently collaboratively accessed, forming a high-relevance data combination, and calculating an IO delay revenue factor, a space cost factor and collaboration access intensity, so as to form a high-relevance data combination; comprehensively evaluating the migration value of each frequent item set data; by constructing a conflict graph model, analyzing a resource competition relationship among different frequent item set data, correcting an expected utility value by using an interference sensing weight, and identifying and avoiding potential resource conflicts; based on the corrected utility value, particle swarm optimization processing is adopted to optimize and select frequent item set data with the highest utility value under the constraint of a high-speed storage space, the frequent item set data are sequentially incorporated into a migration list, and it is ensured that the overall revenue is maximized under limited resources; and according to the priority order of the candidate list, the data block migration is executed by adopting a batch migration strategy, and the storage performance is improved while the system stability is ensured.

Description

Technical Field

[0001] This invention relates to the field of cloud computing data storage optimization, and in particular to a data block association data storage method for improving network security availability. Background Technology

[0002] In the digital economy era, information security is no longer an isolated defensive issue, but a systemic project deeply intertwined with large-scale data storage and efficient transmission. With the deepening of enterprise digital transformation and the widespread adoption of cloud computing technology, the network scale of modern data centers is growing exponentially, and the migration and synchronization of massive amounts of data across data centers and regions has become commonplace. This data is both a core asset and constitutes a huge security attack surface.

[0003] Traditional network security methods often focus on access control of the storage itself. However, data block storage not only affects storage efficiency but also security. Specifically, in the event of a systemic failure, the distributed storage of massive amounts of data blocks (during migration) impacts the overall system performance and rapid recovery. Research has found that traditional dynamic data block storage layout optimization methods are primarily based on data access frequency and correlation analysis. Early research analyzed historical access records of data blocks to identify frequently co-occurring data block combinations and placed highly correlated data block sets in high-speed storage devices. These methods improved data locality to some extent, reduced disk seek time, and thus improved storage system performance. However, existing technologies have gradually revealed the following technical shortcomings when dealing with complex and varied real-world workloads:

[0004] First, traditional methods lack effective analysis and understanding of resource contention between data blocks. In real-world enterprise storage environments, multiple related data block combinations may simultaneously compete for limited I / O channels, cache resources, and controller processing power. Existing solutions only consider the positive correlation between data blocks, ignoring the negative impact of this competition on system performance. When multiple highly correlated data block combinations are migrated to high-speed storage simultaneously, performance may actually degrade due to resource contention, which contradicts the original intention of data layout optimization.

[0005] Secondly, existing technologies lack continuous monitoring and evaluation of the long-term performance status of data blocks. Traditional solutions mainly rely on static access frequency and correlation strength for decision-making, failing to identify the "performance fatigue" phenomenon, i.e., performance instability, that occurs when data blocks are under continuous contention pressure. Specifically, a data block that has been in a high-contention environment for a long time, even if its correlation index is high, may lose its migration value due to continuous performance degradation, and existing methods cannot effectively capture and quantify this dynamic performance change trend.

[0006] Third, current data layout optimization methods do not adequately consider the stability of access patterns. In real-world production environments, data block access patterns often exhibit complex temporal characteristics, including periodic fluctuations, burst access, and random access. Traditional solutions use aggregated statistical metrics (such as average access frequency) that mask these important temporal characteristics, leading to a significant decrease in the effectiveness of optimization decisions when facing unstable workloads.

[0007] Furthermore, existing optimization algorithms have limitations in handling multi-objective trade-offs. The finite nature of high-speed storage space requires solutions to strike a balance between performance gains, space costs, and access stability, while traditional methods typically employ simple linear weighting or threshold judgments, making it difficult to find the globally optimal solution under complex constraints.

[0008] Finally, traditional data migration processes lack sufficient robustness considerations. In large-scale storage systems, data migration operations themselves consume system resources and may interfere with running business operations. Existing solutions often employ aggressive migration strategies and lack effective mechanisms to ensure system stability during the migration process. Summary of the Invention

[0009] The purpose of this invention is to provide a data block association data storage method for improving network security availability, thereby solving the aforementioned technical problems pointed out in the prior art.

[0010] This invention provides a method for storing data block associations to improve network security availability, comprising the following steps:

[0011] Multiple IO requests from the server terminal are collected at preset time periods to obtain an IO request data set; the IO request data set is preprocessed to obtain preprocessed IO requests; the data block combination corresponding to the preprocessed IO requests is obtained; the data block combination corresponding to the preprocessed IO requests is processed through association analysis to obtain multiple frequent itemset data, where each frequent itemset data includes a list of data blocks, support, lift, and total itemset capacity.

[0012] Obtain the data block read / write frequency and storage medium performance parameters corresponding to frequent itemset data; calculate the IO latency benefit factor based on the data block read / write frequency and the storage medium latency difference calculated based on the storage medium performance parameters; calculate the space cost factor based on the total itemset capacity and the preset high-speed storage medium allocation threshold; calculate the cooperative access intensity based on support and lift; and calculate the expected utility value by combining the IO latency benefit factor, space cost factor, and cooperative access intensity.

[0013] Construct a conflict graph model with frequent itemset data as vertices based on all IO requests; calculate the weights of edges in the conflict graph model based on the competition data among all IO requests; calculate the disturbance-aware weight for each frequent itemset data based on the conflict graph model; and use the disturbance-aware weight to correct the expected utility value to obtain the corrected expected utility value.

[0014] Based on the corrected expected utility values, frequent itemset data are analyzed using particle swarm optimization to output a candidate list of data migrations.

[0015] Perform data block migration processing based on the data migration candidate list.

[0016] Preferably, a conflict graph model is constructed based on all IO requests, with frequent itemset data as vertices; the weights of edges in the conflict graph model are calculated based on the competition data among all IO requests; the disturbance-aware weights for each frequent itemset data are calculated based on the conflict graph model; and the expected utility value is corrected using the disturbance-aware weights to obtain the corrected expected utility value, including the following steps:

[0017] Traverse the frequent itemset data involved in each IO request, and determine whether any two frequent itemsets appear in the same IO request or whether the data blocks involved in the two frequent itemsets are accessed simultaneously within the same time window. If any of the above conditions are met, add a competing edge between the corresponding two frequent itemsets. Build a conflict graph model based on all frequent itemsets and competing edges.

[0018] The weights of each competing edge in the conflict graph model are calculated by weighted summation based on the frequency of occurrence of each pair of frequent itemsets in the same IO request, the frequency of occurrence in the same time window, and the proportion of time occupied by two frequent itemsets simultaneously in the IO request queue.

[0019] By utilizing the edge weights of the conflict graph model and combining them with fatigue analysis of competition interference, the interference perception weight of each frequent itemset is calculated. The interference perception weight represents the degree of interference experienced by the frequent itemset in a competitive environment or its ability to interfere with other frequent itemsets, which is used to subsequently correct the expected utility value.

[0020] The expected utility value is updated based on the interference-aware weights to obtain the corrected expected utility value.

[0021] A preferred approach is to represent the conflict graph model as: G=(V,E), where V is the set of vertices of the frequent itemset data and E is the set of competing edges.

[0022] Preferably, the interference-perceived weight of each frequent itemset is calculated using the edge weights of the conflict graph model, combined with fatigue analysis of competition interference, including the following steps:

[0023] For each vertex in the conflict graph model, the sum of the weights of all connecting edges of each vertex is calculated to obtain the basic disturbance intensity.

[0024] Taking each vertex in the conflict graph model as the center, calculate the sum of the basic interference intensity of all adjacent vertices within the preset propagation depth range to obtain the propagation range intensity.

[0025] The frequency sequence of each frequent itemset data in each IO request within a continuous time window is obtained. The standard deviation of the frequency sequence is calculated to obtain the time stability index. The fatigue accumulation rate is calculated by weighting and summing the time stability index, propagation range intensity, and basic interference intensity.

[0026] Extract the baseline performance parameters, safety factor, and real-time load rate of each frequent itemset data; calculate the dynamic fatigue trigger threshold based on the baseline performance parameters, safety factor, and real-time load rate; use the dynamic fatigue trigger threshold and fatigue accumulation rate to determine the fatigue state level of each frequent itemset data; the fatigue state level includes normal state, warning state, and fatigue state.

[0027] The target interference perception weight of each frequent itemset data is obtained by analyzing and calculating the fatigue state level, the basic interference intensity, and the edge weight.

[0028] Preferably, the target interference perception weight of each frequent itemset data is obtained by analyzing and calculating based on the fatigue state level, the basic interference intensity, and the edge weights, including the following steps:

[0029] The process involves: calculating the fatigue duration of each frequent itemset data point in a fatigued state within the collection period of an IO request; identifying frequent itemsets data whose fatigue duration exceeds a threshold as target fatigue frequent itemsets data; obtaining the propagation impact frequent itemsets data within a preset propagation depth range for each target fatigue frequent itemsets data; obtaining the current performance parameters of each propagation impact frequent itemsets data; calculating the performance degradation ratio of each propagation impact frequent itemsets data based on the current performance parameters and baseline performance parameters; and calculating the fatigue impact range value of each target fatigue frequent itemsets data using the performance degradation ratio and the edge weights between the target fatigue frequent itemsets data and each propagation impact frequent itemsets data.

[0030] The fatigue impact range value of each target fatigue frequent itemset data is used as the original disturbance perception weight. A fixed fatigue state coefficient is set for all frequent itemset data other than the target fatigue frequent itemset data using the ratio of fatigue accumulation rate to dynamic fatigue trigger threshold. The original disturbance perception weight of each frequent itemset data other than the target fatigue frequent itemset data is calculated by multiplying the fixed fatigue state coefficient by the basic disturbance pre-degree of each frequent itemset data other than the target fatigue frequent itemset data. All original disturbance perception weights are normalized to obtain the target disturbance perception weight of each frequent itemset data.

[0031] Preferably, based on the corrected expected utility value, frequent itemset data is analyzed using particle swarm optimization to output a data migration candidate list, including the following steps:

[0032] Initialize the particle swarm parameters, which include the iteration counter and the maximum number of iterations threshold, the initial particle swarm, the initial learning rate, and the initial particle search speed.

[0033] The iteration counter is initially set to 0; the initial particle swarm consists of N particles, each particle represents a candidate migration list, the vector length of each particle represents the number of frequent itemset data, and the position of each particle vector represents whether the frequent itemset data is migrated or not.

[0034] Based on the frequency of IO requests to the frequent itemset data corresponding to each particle, an access fluctuation analysis is performed to obtain the fluctuation coefficient of the frequent itemset data.

[0035] Migration fitness is calculated using the volatility coefficient of frequent itemset data, the corrected expected utility value, and the total itemset capacity of each frequent itemset data.

[0036] The target data migration candidate list is obtained by iteratively judging and updating the output based on the migration fitness and the iteration counter.

[0037] Preferably, the target data migration candidate list is obtained by iteratively judging and updating the output based on the migration fitness and the iteration counter, including the following steps:

[0038] Determine if it is the first iteration. If yes, take the current migration fitness of each particle as its individual optimal fitness, and select the migration fitness with the largest value among all migration fitness values ​​as the global optimal fitness. If it is not the first iteration, increment the iteration count by 1 to get the current iteration count. Compare the migration fitness of each particle with its individual optimal fitness value at the current iteration count, and select the larger migration fitness value as the new individual optimal fitness value. Select the individual optimal fitness value with the largest value among all individual optimal fitness values ​​of all particles as the new global optimal fitness value at the current iteration count.

[0039] Determine if the current iteration number is greater than or equal to the maximum iteration number threshold. If yes, output the particle corresponding to the new global optimal fitness at the current iteration number as the target data migration candidate list. If no, calculate if the increase ratio of the new global optimal fitness at the current iteration number to the global optimal fitness at the previous iteration number is less than or equal to the increase ratio threshold. If yes, output the particle corresponding to the new global optimal fitness at the current iteration number as the target data migration candidate list. If no, update the initial particle search velocity using the individual optimal fitness at the current iteration number, the global optimal fitness, and the learning rate to obtain a new initial particle search velocity. Update the initial particle swarm using the new particle search velocity to obtain a new initial particle swarm. Return the new initial particle swarm and the new particle search velocity to the above steps for re-iteration until the target data migration candidate list is output.

[0040] Preferably, access fluctuation analysis is performed based on the I / O request frequency of the frequent itemset data corresponding to each particle to obtain the fluctuation coefficient of the frequent itemset data, including the following steps:

[0041] The frequency of IO requests for each frequent itemset data at consecutive time points is extracted from the preprocessed IO requests. Based on the frequency of IO requests at consecutive time points, a three-dimensional feature tensor T[k][i][j] is constructed at multiple time scales; where k represents the time point index, i represents the frequent itemset data index, and j represents the time scale index. Each three-dimensional feature tensor T[k][i][j] represents the frequency of IO requests for the i-th frequent itemset data at the k-th time point at the j-th time scale.

[0042] Based on the three-dimensional feature tensor, calculate the autocorrelation fluctuation characteristics of the i-th frequent itemset data at consecutive time points under the j-th time scale;

[0043] At a fixed time point k, calculate the instantaneous coherence of any two frequent itemsets at the same time scale. Average the instantaneous coherence across all time points to obtain the coherent fluctuation characteristics of each pair of frequent itemsets. For each frequent itemset, use the coherent fluctuation characteristics to calculate its average coherent fluctuation characteristics with all other frequent itemsets.

[0044] The volatility coefficient of frequent itemset data is calculated by weighted summation based on the autocorrelation volatility characteristics and the average co-variance volatility characteristics.

[0045] Preferably, the calculation method of the autocorrelation fluctuation characteristics is as follows: for each frequent itemset in the complete time series at each time scale, by selecting two different time points for comparative analysis, the autocorrelation contribution value between the two time points is obtained. Then, by iterating through the autocorrelation contribution values ​​of every two time points in the complete time series, the autocorrelation fluctuation characteristics are obtained.

[0046] In another aspect, the present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for data block association data storage to improve network security availability.

[0047] Compared with the prior art, the embodiments of the present invention have at least the following technical advantages:

[0048] Analysis of the data block association data storage method for improving network security availability provided by this invention reveals that, in practical applications, firstly, by analyzing historical IO requests, frequently accessed data block sets are identified, forming highly associated data combinations. Then, by calculating IO latency benefit factors, space cost factors, and collaborative access intensity, the migration value of each frequent itemset is comprehensively evaluated. Further, by constructing a conflict graph model, the resource competition relationship between different frequent itemsets is analyzed, and the expected utility value is corrected using interference-aware weights to identify and avoid potential resource conflicts. Further still, based on the corrected utility value, particle swarm optimization is used to optimize and select the most frequent itemset data with the highest utility value under high-speed storage space constraints, and these are sequentially included in the migration list to ensure maximum overall benefit with limited resources. Finally, according to the priority order of the candidate list, a batch migration strategy is used to execute data block migration, improving storage performance while ensuring system stability. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the main process of a data block-related data storage method for improving network security availability;

[0050] Figure 2This is a schematic diagram simulating frequent itemset data in a data block-related data storage method for improving network security availability;

[0051] Figure 3 This is a schematic diagram simulating a conflict graph model in a data block association data storage method for improving network security availability;

[0052] Figure 4 This is a schematic diagram simulating the propagation depth range in a data block-related data storage method for improving network security availability. Detailed Implementation

[0053] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0055] Example 1

[0056] like Figure 1 As shown, Embodiment 1 of the present invention provides a data block association data storage method for improving network security availability, including the following operation steps:

[0057] Step S10: Collect multiple IO requests from the server terminal at a preset time period to obtain an IO request data set; preprocess the IO request data set to obtain preprocessed IO requests; obtain the data block combination corresponding to the preprocessed IO requests; process the data block combination corresponding to the preprocessed IO requests through association analysis to obtain multiple frequent itemset data, wherein each frequent itemset data includes a data block list, support, lift, and total itemset capacity.

[0058] It should be noted that: such as Figure 2 As shown, the above-described embodiment of this application identifies frequently accessed data block combinations by analyzing the historical access records of the storage system. Specifically, it continuously records all IO requests from the server terminal at a preset time period. Each IO request includes the operation time, source virtual machine identifier, operation type, and a list of data blocks involved. These IO requests are grouped by time window to form an IO request data set. Then, the IO request data set is cleaned to remove abnormal data block access records and the data block addresses are uniformly formatted.

[0059] Next, the FP-Growth algorithm is used to mine co-occurrence relationships between data blocks. This algorithm first constructs a frequent pattern tree to compress and store the co-occurrence information of all data blocks. Then, it recursively mines the conditional pattern base to find all data block combinations that meet the minimum support threshold. For each identified data block combination, its complete data block list is recorded, and three key metrics are calculated: support (the frequency of occurrence of the data block combination in all IO requests), lift (the strength of the association between data blocks in the data block group), and total itemset capacity (the total itemset capacity is the set capacity of frequent itemset data). For example, after analyzing 1000 IO requests, it was found that data blocks A and B appeared simultaneously in 800 IO requests, with a support of 80%, a lift of 2.5, and a total capacity of 16MB.

[0060] Step S20: Obtain the data block read / write frequency and storage medium performance parameters corresponding to frequent itemset data; calculate the IO latency benefit factor based on the data block read / write frequency and the storage medium latency difference calculated based on the storage medium performance parameters; calculate the space cost factor based on the total itemset capacity and the preset high-speed storage medium allocation threshold; calculate the cooperative access strength based on support and lift; calculate the expected utility value by combining the IO latency benefit factor, space cost factor, and cooperative access strength; the storage medium performance parameters include high-speed storage medium access latency indicators and low-speed storage medium access latency indicators.

[0061] It should be noted that the above embodiments of this application establish a multi-dimensional evaluation system to quantify the expected benefits of migrating each frequent itemset data to high-speed storage (i.e., the expected utility value mentioned above, which initially quantifies the theoretical performance benefits that can be brought by migrating a frequent itemset data (a set of associated data blocks) from low-speed storage (such as HDD) to high-speed storage (such as SSD). Specifically, the storage system monitoring module obtains the read and write frequency parameters of each data block corresponding to the frequent itemset data, including read IOPS (IOPS is a performance indicator that measures the number of read and write operations processed per second by a computer storage device, and is often used to evaluate the random access performance of devices such as hard disk drives (HDD) and solid-state drives (SSD)) and write IOPS; at the same time, the storage medium performance parameters are obtained, where the storage medium latency difference refers to the difference in the average time required for high-speed storage medium (such as SSD) and low-speed storage medium (such as HDD) to process the same IO request. For example, if the access latency of SSD is 0.5ms and the access latency of HDD is 5ms, then the storage medium latency difference is 4.5ms.

[0062] Then, the IO latency benefit factor is calculated based on the data block read / write frequency and storage medium latency difference: read IOPS multiplied by read latency difference, plus write IOPS multiplied by write latency difference. For example, if a frequent itemset contains data blocks with a total read IOPS of 1000, write IOPS of 500, and a storage medium latency difference of 4.5ms, then the IO latency benefit factor is 1000 × 4.5ms + 500 × 4.5ms = 6750. Further, the space cost factor is calculated based on the total itemset capacity and the high-speed storage medium allocation threshold, where the high-speed storage medium allocation threshold refers to the maximum available space of high-speed storage medium reserved for data layout optimization, used to prevent optimization... The process exhausts all high-speed storage resources. If the total itemset capacity is 16MB, the total high-speed storage capacity is 250GB, and the allocation threshold is 80%, then the maximum available space for migration is 250GB × 80% = 200GB. The space cost factor is the total itemset capacity / available space under the allocation threshold, i.e., 16MB / 200GB = 0.00008. Furthermore, the collaborative access strength is calculated based on support and lift, using the geometric mean formula: the square root of support multiplied by the logarithm of lift. For example, for an itemset with 80% support and 2.5 lift, the collaborative access strength is sqrt(0.8) × log(2.5) ≈ 0.894 × 0.916 = 0.819.

[0063] Finally, the expected utility value is calculated by combining three factors: the IO latency benefit factor multiplied by the collaborative access intensity, and then divided by the space cost factor. Using the example above, the expected utility value is (6750×0.819) / 0.00008≈69,100,000. All calculation parameters are dynamically adjusted through the configuration management module to ensure that the evaluation results conform to the characteristics of the actual operating environment.

[0064] Step S30: Construct a conflict graph model with frequent itemset data as vertices based on all IO requests; calculate the weights of edges in the conflict graph model based on the competition data between all IO requests; calculate the interference-aware weight of each frequent itemset data based on the conflict graph model; use the interference-aware weight to correct the expected utility value to obtain the corrected expected utility value.

[0065] Step S40: Based on the corrected expected utility value, analyze the frequent itemset data through particle swarm optimization and output a data migration candidate list.

[0066] Step S50: Perform data block migration processing based on the data migration candidate list.

[0067] It should be noted that the technical solution adopted in the above-described embodiments of this application first prepares the migration execution environment according to the data migration candidate list, creates a snapshot of the current data layout, establishes a migration transaction log, and prepares the data structures and resources required for rollback. Then, according to the priority of the candidate list, the migration task is divided into multiple batches, each batch containing 4-6 frequent itemsets of data, with an estimated execution time of 2-3 hours, and a 30-minute interval between batches for system stability checks. Then, the data block migration operation is performed in batches. When each data block is migrated, space is first allocated in the target storage, and then data replication is started. During replication, the write operation of the data block is applied to both the source storage and the target storage. For example, the first batch migrates 5 frequent itemsets of data with a total capacity of 64GB, using a parallel transmission method, opening 4 transmission channels simultaneously, with each channel's rate limited to 200MB / s.

[0068] The above-described embodiment first analyzes historical IO requests in step S10 to identify frequently accessed data block sets, forming highly correlated data combinations. Then, in step S20, it calculates the IO latency benefit factor, space cost factor, and collaborative access intensity to comprehensively evaluate the migration value of each frequent itemset. Further, in step S30, it constructs a conflict graph model to analyze the resource competition relationship between different frequent itemsets, and uses interference-aware weights to correct the expected utility value, identifying and avoiding potential resource conflicts. Further, in step S40, based on the corrected utility value, it uses particle swarm optimization to optimize and select the most frequent itemset data with the highest utility value under high-speed storage space constraints, and includes it in the migration list in sequence to ensure that the overall benefit is maximized under limited resources. Finally, in step S50, it performs data block migration according to the priority order of the candidate list, using a batch migration strategy to improve storage performance while ensuring system stability.

[0069] Specifically, in step S30, a conflict graph model with frequent itemset data as vertices is constructed based on all IO requests; the weights of edges in the conflict graph model are calculated based on the competition data among all IO requests; the interference-aware weight of each frequent itemset data is calculated based on the conflict graph model; and the expected utility value is corrected using the interference-aware weight to obtain the corrected expected utility value, including the following steps:

[0070] Step S31: Traverse the frequent itemset data involved in each IO request, and determine whether any two frequent itemset data appear in the same IO request or whether the data blocks involved in the two frequent itemset data are accessed at the same time within the same time window. If any of the above judgment conditions are met, add a competing edge between the corresponding two frequent itemset data; build a conflict graph model based on all frequent itemset data and competing edges.

[0071] It should be noted that in the above embodiments of this application, by traversing each IO request and extracting all frequent itemset data involved in the request, for any two frequent itemset data, if they appear in the same IO request, or their data blocks are accessed within the same time window (such as the preset time period in step S10), then an edge (i.e., the aforementioned competing edge) is added between them, indicating that the two frequent itemset data have a competitive relationship, such as... Figure 3 As shown in the diagram, the red values ​​of 0.8, 0.6, 0.7, 0.5, 0.4, and 0.3 represent the edge weights of each competing variable.

[0072] The conflict graph model is represented as: G=(V,E), where V is the set of vertices of frequent itemset data (the number of vertices is the total number of frequent itemset data), and E is the set of competing edges. Each edge connects vertices c and d, indicating that there is competition between frequent itemsets c and d.

[0073] The above-described embodiments of this application analyze the co-occurrence and temporal overlap of IO requests to identify potential resource contention between frequent itemset data, providing a basic structure for subsequent calculation of contention intensity; each vertex in the vertex set V corresponds to a frequent itemset data, including a list of data blocks, support, lift, and total itemset capacity (obtained from step S10); the initial establishment of the edge set E is based on the overlapping relationship of IO requests, ensuring that the conflict graph model can truly reflect the resource contention pattern in the storage system.

[0074] Step S32: Based on the frequency of occurrence of each pair of frequent itemsets in the same IO request (the proportion of the number of times frequent itemsets c and d appear together in the same IO request to the total number of IO requests), the frequency of occurrence in the same time window (the proportion of the number of times frequent itemsets c and d are accessed in the same time window to the total number of time windows), and the proportion of time occupied by two frequent itemsets when they appear in the IO request queue at the same time (calculated based on the waiting queue length and resource occupation time of the IO request, reflecting the degree of competition for storage resources (such as bandwidth and IOPS) when two frequent itemsets are accessed at the same time), the weight of each competing edge in the conflict graph model is calculated by weighted summation.

[0075] It should be noted that in the above embodiments of this application, by quantifying competitive data, the abstract competitive relationship is transformed into specific edge weights, so that the conflict graph model is transformed from a structural graph into a weighted graph; the calculation of edge weights is closely dependent on the historical behavior of IO requests, ensuring the objectivity of the competition intensity assessment.

[0076] Step S33: Calculate the disturbance perception weight of each frequent itemset (vertex) using the edge weights of the conflict graph model and the fatigue analysis of competition interference. The disturbance perception weight represents the degree of interference experienced by the frequent itemset in the competitive environment or its ability to interfere with other frequent itemsets, which is used to subsequently correct the expected utility value.

[0077] Step S34: Update the expected utility value based on the interference perception weight to obtain the corrected expected utility value;

[0078] ;in, η is the interference correction coefficient, which controls the intensity of the influence of interference factors on the utility value; the larger η is, the more significant the correction effect of the interference perception weight.

[0079] It should be noted that the above-described embodiments of this application identify the competitive relationships between frequent itemsets by traversing each IO request. Specifically, they check whether any two frequent itemsets appear in the same IO request or whether their related data blocks are accessed simultaneously within the same time window. If either condition is met, a competing edge is added between the corresponding two frequent itemsets, establishing a conflict graph model. This ensures that the conflict graph model can accurately reflect the resource contention patterns in the storage system, providing a foundation for subsequent calculations of competition intensity. By analyzing the co-occurrence and temporal overlap of IO requests, potential resource competition is initially identified. Furthermore, based on historical data from all IO requests, the competition intensity between any two frequent itemsets is quantified, transforming the abstract competition... Relationships are transformed into specific edge weights, converting the conflict graph model from a structural graph into a weighted graph. This objectively assesses the intensity of competition (such as the degree of competition for storage resources like bandwidth and IOPS), providing data support for subsequent interference analysis. Then, using the edge weights in the conflict graph model, combined with fatigue analysis of competition interference, the interference-aware weight of each frequent itemset (vertex) is calculated. This weight represents the degree of interference experienced by the frequent itemset in a competitive environment or its ability to interfere with other frequent itemsets. This transforms competitive relationships into interference indicators for individual frequent itemsets, laying the foundation for revising the expected utility value. Finally, the expected utility value is updated based on the interference-aware weights to obtain the revised expected utility value, improving the accuracy of utility value assessment and assisting in optimizing storage resource allocation.

[0080] Specifically, in step S33, the interference-aware weight of each frequent itemset data (vertices) is calculated using the edge weights of the conflict graph model and in conjunction with fatigue analysis of competition interference, including the following steps:

[0081] Step S331: For each vertex in the conflict graph model, calculate the sum of the weights of all connecting edges of each vertex to obtain the basic interference intensity;

[0082] The basic disturbance intensity reflects the total competitive pressure between the vertex and its immediate neighboring vertices;

[0083] In the specific calculation, the basic disturbance intensity of all vertices is normalized, and the values ​​are mapped to the range of 0-1 to facilitate subsequent comparison and analysis.

[0084] It should be noted that: the above embodiments of this application establish a basic interference intensity benchmark by aggregating direct competitive relationships; normalization processing eliminates dimensional differences, providing a unified comparison basis for subsequent multi-dimensional analysis.

[0085] Step S332: Taking each vertex in the conflict graph model as the center, calculate the sum of the basic interference intensity of all adjacent vertices within the preset propagation depth range to obtain the propagation range intensity;

[0086] It should be noted that in the above embodiments of this application, the preset propagation depth range is a fixed empirical parameter preset based on a trade-off between system architecture, performance requirements, and computational complexity. It represents the spatial breadth and depth of the resource contention interference experienced by a frequent itemset data (vertices) in the conflict graph topology. For example: Figure 4 As shown, the current server has four frequent itemset data: A, B, C, and D. Among them, frequent itemset data A is the frequent itemset data of user authentication service data, frequent itemset data B is the frequent itemset data of order query service data, frequent itemset data C is the frequent itemset data of product information service data, and frequent itemset data D is the frequent itemset data of log recording service data.

[0087] In this scenario, frequent itemset data A and B are frequently accessed simultaneously when users log in and query orders (direct competition, with one edge). Frequent itemset data B and C are frequently accessed simultaneously when displaying order details (direct competition, with one edge), while frequent itemset data C and D have a weaker correlation. When frequent itemset data A is under extremely high load and access to it is very slow, frequent itemset data B is directly affected because it directly competes for resources with A, thus slowing down access to B. In this case, the propagation depth is 1. For example, if the interference from frequent itemset data A propagates to frequent itemset data B, and B's performance degrades, when B and C are accessed together, B cannot respond quickly, thus affecting performance. The pressure is passed to frequent itemset data C. At this point, although frequent itemset data A and frequent itemset data C are not directly related, frequent itemset data C is also indirectly affected by frequent itemset data A. Therefore, the propagation range of the interference from frequent itemset data A covers frequent itemset data C, and the propagation depth range is 2. In this embodiment, when calculating the propagation range of the interference from frequent itemset data A (assuming the propagation depth range = 2), the interference intensity of frequent itemset data B and frequent itemset data C is also taken into consideration. This allows the server to recognize that migrating frequent itemset data A to high-speed storage will not only make frequent itemset data A and frequent itemset data B faster, but also benefit the indirectly affected frequent itemset data C, thereby significantly improving the expected utility value of frequent itemset data A and increasing its priority in the migration candidate list. Typically, the propagation depth range is 2.

[0088] Step S333: Obtain the frequency sequence of each frequent itemset data in each IO request within a continuous time window, and use the frequency sequence to calculate the standard deviation to obtain the time stability index; use the time stability index, propagation range intensity and basic interference intensity to perform weighted summation to calculate the fatigue accumulation rate;

[0089] It should be noted that in the above embodiments of this application, the fatigue accumulation rate characterizes the overall speed and stability of the performance degradation of a frequent itemset data. A high fatigue accumulation rate means that the frequent itemset data is continuously and steadily sliding towards the performance bottleneck.

[0090] Step S334: Extract the baseline performance parameters, safety factor values, and real-time load rate of each frequent itemset data; calculate the dynamic fatigue trigger threshold based on the baseline performance parameters, safety factor values, and real-time load rate; use the dynamic fatigue trigger threshold and fatigue accumulation rate to determine the fatigue state level of each frequent itemset data; the fatigue state level includes normal state, warning state, and fatigue state.

[0091] It should be noted that in the above embodiments of this application, the benchmark performance parameters include the average response time benchmark value, the IOPS benchmark value, and the throughput benchmark value. These are obtained by statistically analyzing historical performance data of each frequent itemset during low-load periods. Then, the average response time benchmark value, the IOPS benchmark value, and the throughput benchmark value are weighted and summed to obtain the benchmark performance parameters, which represent the best performance that a certain frequent itemset (a set of data blocks) can achieve in an ideal environment with no resource contention and extremely low system load. The above-mentioned safety factor value is preset by the system administrator according to the business tolerance and system stability requirements, and is usually 0.7. The real-time load rate includes CPU utilization, memory utilization, and IO subsystem utilization. The three are weighted and summed to obtain the real-time load rate, which reflects the current busy level of the server. Based on the above-mentioned benchmark performance parameters, safety factor value, and real-time load rate, the dynamic fatigue trigger threshold is calculated, that is, dynamic fatigue trigger threshold = benchmark performance parameters × safety factor value × (1 - real-time load rate).

[0092] The above-mentioned fatigue state level is determined as follows: if the fatigue accumulation rate is less than 0.7 × dynamic fatigue trigger threshold, the frequent itemset data is judged to be in a normal state; if 0.7 × dynamic fatigue trigger threshold is less than or equal to the fatigue accumulation rate and the dynamic fatigue trigger threshold, the frequent itemset data is judged to be in a warning state; if the fatigue accumulation rate is greater than or equal to the dynamic fatigue trigger threshold, the frequent itemset data is judged to be in a fatigue state.

[0093] Step S335: Calculate the fatigue duration of each frequent itemset data in a fatigued state within the collection time period of the IO request; determine the frequent itemset data whose fatigue duration exceeds the fatigue duration threshold as target fatigue frequent itemset data; obtain the propagation impact frequent itemset data within the preset propagation depth range of each target fatigue frequent itemset data; obtain the current performance parameter index of each propagation impact frequent itemset data; calculate the performance decay ratio of each propagation impact frequent itemset data based on the current performance parameter index and the benchmark performance parameter index (performance decay ratio = (benchmark performance parameter index - current performance parameter index) / benchmark performance parameter index); calculate the fatigue impact range value of each target fatigue frequent itemset data using the performance decay ratio and the edge weights between the target fatigue frequent itemset data and each propagation impact frequent itemset data (i.e., multiply the performance decay ratio by the edge weights and then normalize).

[0094] It should be noted that in the above embodiments of this application, the fatigue influence range value of each target fatigue frequent itemset data quantifies the degree of fatigue influence of the fatigue state of each target fatigue frequent itemset data on the fatigue of each frequent itemset data within its preset propagation depth range.

[0095] Step S336: Based on the fatigue influence range value of each target fatigue frequent itemset data as the original disturbance perception weight, set a fixed fatigue state coefficient for all frequent itemset data other than the target fatigue frequent itemset data using the ratio of fatigue accumulation rate to dynamic fatigue trigger threshold; multiply the fixed fatigue state coefficient by the basic disturbance pre-degree of each frequent itemset data other than the target fatigue frequent itemset data to calculate the original disturbance perception weight of each frequent itemset data other than the target fatigue frequent itemset data; normalize all the original disturbance perception weights to obtain the target disturbance perception weight of each frequent itemset data.

[0096] It should be noted that in the above embodiments of this application, a fixed fatigue state coefficient is set for all frequent itemset data other than the target fatigue frequent itemset data using the ratio of fatigue accumulation rate to dynamic fatigue trigger threshold. For example, when the ratio of fatigue accumulation rate to dynamic fatigue trigger threshold is < 0.7 (normal state), the fixed fatigue state coefficient takes a low value, such as 0.2, indicating almost no fatigue; when 0.7 ≤ the ratio of fatigue accumulation rate to dynamic fatigue trigger threshold is < 1.0 (warning state), the fixed fatigue state coefficient takes a medium value, such as 0.5, indicating signs of fatigue but not yet triggered; when the ratio of fatigue accumulation rate to dynamic fatigue trigger threshold is ≥ 1.0 (fatigue state but short duration), the fixed fatigue state coefficient takes a high value, such as 0.8, indicating fatigue, but the duration has not yet reached the standard.

[0097] In the embodiments described above, firstly, for each vertex in the conflict graph model, the sum of the weights of all its connecting edges is calculated to obtain the basic interference intensity, reflecting the total competitive pressure between the vertex and its directly adjacent vertices. By normalizing the basic interference intensity of all vertices, the values ​​are mapped to the range of 0-1, eliminating dimensional differences and providing a unified basis for comparison in subsequent multi-dimensional analysis. Further, with each vertex as the center, the sum of the basic interference intensity of all adjacent vertices within a preset propagation depth range is calculated to obtain the propagation range intensity, identifying the breadth and depth of the impact of resource competition interference on frequent itemset data in the conflict graph topology. By using the preset propagation depth range, indirect competitive relationships are taken into consideration, allowing the server to assess the propagation effect of the interference. Further, by obtaining the frequency sequence of each frequent itemset data within a continuous time window, its standard deviation is calculated to obtain a time stability index. This index is then weighted and summed with the propagation range intensity and the basic interference intensity to obtain the fatigue accumulation rate frequency. The comprehensive speed and stability of performance degradation of frequent itemset data provide a quantitative basis for subsequent fatigue state determination. Furthermore, a dynamic fatigue trigger threshold is calculated based on baseline performance parameters, safety factor values, and real-time load rates. Using this threshold and the fatigue accumulation rate, the fatigue state level (normal state, warning state, fatigue state) of each frequent itemset is determined. The dynamic fatigue trigger threshold enables state determination adapted to the real-time load of the system, combining performance benchmarks, system safety requirements, and current load conditions to ensure the accuracy and real-time nature of state determination. Further, by statistically analyzing the fatigue duration of each frequent itemset, the target fatigue frequent itemset is identified. Frequent itemset data within its propagation range are obtained, and the fatigue influence range value is calculated based on the performance degradation ratio and edge weights. This quantifies the fatigue impact of each target fatigue frequent itemset on other frequent itemset data within its propagation range. The combined calculation of the performance degradation ratio and edge weights accurately reflects the diffusion effect of the fatigue state.

[0098] Furthermore, the fatigue impact range value of the target fatigue frequent itemset data is used as its original disturbance perception weight; for other frequent itemset data, the original disturbance perception weight is obtained by multiplying the fixed fatigue state coefficient (based on the ratio of fatigue accumulation rate to dynamic fatigue trigger threshold) with its basic disturbance intensity; finally, all original weights are normalized, and the disturbance perception weight of all frequent itemset data is calculated uniformly. By distinguishing the processing methods of the target fatigue itemset and other itemsets, and combining the fixed fatigue state coefficient, it is ensured that the weight calculation considers both the current fatigue state and the potential fatigue risk, providing accurate input for subsequent utility value correction.

[0099] Specifically, in step S40, based on the corrected expected utility value, the frequent itemset data is analyzed through particle swarm optimization to output a data migration candidate list, including the following steps:

[0100] Step S41: Initialize particle swarm parameters, which include an iteration counter and a maximum iteration threshold, an initial particle swarm, an initial learning rate, and an initial particle search speed.

[0101] The iteration counter is initially set to 0; the initial particle swarm consists of N particles, each particle represents a candidate migration list, the vector length of each particle represents the number of frequent itemset data, and the position of each particle vector represents whether the frequent itemset data is migrated or not.

[0102] It should be noted that in the above embodiments of this application, the particle swarm is initialized based on frequent itemset data as the basis for subsequent optimization.

[0103] Step S42: Perform access fluctuation analysis based on the IO request frequency of the frequent itemset data corresponding to each particle to obtain the fluctuation coefficient of the frequent itemset data;

[0104] Step S43: Calculate the migration fitness using the volatility coefficient of frequent itemset data, the corrected expected utility value, and the total itemset capacity of each frequent itemset data.

[0105] The migration fitness is calculated as follows:

[0106] ;

[0107] In the formula, Let be the corrected expected utility value of the i-th frequent itemset data; Let be the volatility coefficient of the i-th frequent itemset data. This represents the value of the nth particle at the mth position (the value is usually 1 or 0, where 1 indicates that the frequent itemset data is migrated and 0 indicates that it is not migrated). This represents the total capacity of the itemsets in the frequent itemset data of the data migration candidate list corresponding to the nth particle; The maximum available space for high-speed storage media. This is the space penalty coefficient;

[0108] It should be noted that in the above embodiments of this application, a multi-dimensional evaluation of the migration scheme is achieved by constructing a composite fitness function, wherein the calculation formula is... The basic return of some calculation schemes, including the volatility coefficient Correction of expected utility value Stability corrections are performed to avoid bias in the evaluation caused by frequently accessed, high-value itemset data with unstable access; the calculation formula... Partially constrained by space limitations, when the total capacity of the migration scheme... No more than the space limit When, this part has a value of 1; when Exceed At that time, this part of the value decreases as the overspending ratio increases, forming a space penalty. The space penalty coefficient γ controls the penalty intensity. The larger the value of γ, the more severe the penalty for space capacity overspending. This ensures that the particle swarm optimization process balances performance gains, access stability and space constraints within a limited high-speed storage space, and outputs a data migration candidate list that maximizes overall benefits.

[0109] For example, if a migration scheme selects three frequent itemsets with modified expected utilities of 8 million, 6 million, and 4 million respectively, and volatility coefficients of 0.9, 0.7, and 0.8 respectively, then the base payoff is 8 million × 0.9 + 6 million × 0.7 + 4 million × 0.8 = 14.6 million. If the total capacity of the scheme is 180GB and does not exceed the 200GB space limit, the fitness is 14.6 million. If the total capacity of the scheme is 240GB and exceeds the space limit by 20%, the fitness at γ=2 is 14.6 million × [1-0.2]^2 = 9.344 million.

[0110] Step S44: Determine if it is the first iteration. If yes, use the current migration fitness of each particle as its individual optimal fitness, and select the migration fitness with the largest migration fitness among all migration fitnesss as the global optimal fitness. If it is not the first iteration, increment the iteration count by 1 to obtain the current iteration count. Compare the migration fitness of each particle with its individual optimal fitness at the current iteration count, and select the larger migration fitness as the new individual optimal fitness (i.e., compare and determine if the migration fitness of the current particle at the current iteration count is greater than the individual optimal fitness of the current particle. If yes, select the migration fitness of the current particle at the current iteration count as the new individual optimal fitness; if no, retain the individual optimal fitness from the previous iteration as the new individual optimal fitness). Select the individual optimal fitness with the largest individual optimal fitness among all particles as the new global optimal fitness at the current iteration count.

[0111] Step S45: Determine whether the current iteration number is greater than or equal to the maximum iteration number threshold. If yes, output the particle corresponding to the new global optimal fitness at the current iteration number as the target data migration candidate list. If no, calculate whether the increase ratio between the new global optimal fitness at the current iteration number and the global optimal fitness at the previous iteration number (the increase ratio is the global optimal fitness at the current iteration minus the global optimal fitness at the previous iteration to obtain the fitness difference, and then using the fitness difference divided by the global optimal fitness at the previous iteration to obtain the increase ratio) is less than or equal to the increase ratio threshold. If yes, output the particle corresponding to the new global optimal fitness at the current iteration number as the target data migration candidate list. If no, update the initial particle search velocity using the individual optimal fitness at the current iteration number, the global optimal fitness, and the learning rate to obtain a new initial particle search velocity. Update the initial particle swarm using the new particle search velocity to obtain a new initial particle swarm. Return the new initial particle swarm and the new particle search velocity to step S41 above for re-iteration until the target data migration candidate list is output.

[0112] It should be noted that the above embodiments of this application first set an iteration counter and a maximum iteration threshold to generate an initial particle swarm containing N particles, define an initial learning rate and particle search speed, and establish a basic framework for the optimization process. Each particle represents a candidate migration list, and its vector position is represented by 1 or 0 to indicate whether to migrate the corresponding frequent itemset data, ensuring that the search space covers all possible migration scheme combinations. Furthermore, based on the IO request frequency of the frequent itemset data corresponding to each particle, access fluctuation analysis is performed to obtain a fluctuation coefficient, quantifying the access stability of the frequent itemset data, providing a correction basis for subsequent fitness calculation, and avoiding the bias of high-value but unstable access data in the evaluation. Furthermore, using the fluctuation coefficient, the corrected expected utility value, and the total itemset capacity, the migration fitness of each particle is calculated through a composite function to achieve multi-dimensional scheme evaluation. Furthermore, a judgment process is performed. If it is the first iteration, the current fitness of each particle is taken as an individual... The optimal fitness of the individual is determined, and the maximum value is selected as the global optimal fitness. If it is not the first iteration, the current fitness is compared with the historical optimal fitness of the individual and updated accordingly. The global optimal fitness is then redefined. The optimal solutions of previous iterations are retained through iterative comparison to ensure that the optimization direction always moves towards fitness improvement, providing guidance for particle swarm evolution. Further time judgment is performed. First, it is determined whether the current iteration count has reached a threshold. If not, the increase ratio of the global optimal fitness is calculated to see if it is less than or equal to the threshold. If either condition is met, the particle corresponding to the current global optimal fitness is output as the final result. Otherwise, the particle search speed and particle swarm position are updated and the iteration is restarted. Two termination mechanisms are provided to ensure optimization efficiency: hard termination based on the number of iterations and soft termination based on the degree of convergence. By dynamically updating the particle search speed (combining the individual optimal and global optimal fitness) and particle swarm position, the search space is continuously explored to ensure that the comprehensive benefits of the output solution are maximized.

[0113] Specifically, in step S42, access fluctuation analysis is performed based on the IO request frequency of the frequent itemset data corresponding to each particle to obtain the fluctuation coefficient of the frequent itemset data, including the following steps:

[0114] Step S421: Extract the IO request frequency of each frequent itemset data at consecutive time points from the preprocessed IO requests. Based on the IO request frequency at consecutive time points, construct a three-dimensional feature tensor T[k][i][j] at multiple time scales; where k represents the time point index (i.e., k=1,2,3...,K, a total of K consecutive sampling time points), i represents the frequent itemset data index (i.e., i=1,2,...I, a total of I frequent itemset data), and j represents the time scale index (i.e., j=1,2,...J, a total of J different time scales). Each three-dimensional feature tensor T[k][i][j] represents the IO request frequency of the i-th frequent itemset data at the k-th time point at the j-th time scale.

[0115] Step S422: Based on the three-dimensional feature tensor, calculate the autocorrelation fluctuation characteristics of the i-th frequent itemset data at consecutive time points on the j-th time scale;

[0116] It should be noted that in the above embodiments of this application, for each frequent itemset, in the complete time series at each time scale, two different time points are selected for comparative analysis to obtain the autocorrelation contribution value between the two time points. Then, by traversing the autocorrelation contribution values ​​of every two time points in the complete time series, the autocorrelation fluctuation characteristics are obtained.

[0117] The calculation method for the aforementioned autocorrelation fluctuation characteristics is as follows:

[0118] ;

[0119] In the above formula, K is the total number of time points in the complete time series. Let be the time decay coefficient, where 'a' is the a-th time point and 'b' is the b-th time point, and the a-th time point is earlier than the b-th time point. Let be the frequency of I / O requests for the i-th frequent itemset data at time point a on the j-th time scale. Let be the frequency of I / O requests for the i-th frequent itemset data at time point b on the j-th time scale. It is a smoothing constant. The autocorrelation contribution between the two time points mentioned above is calculated using two summations. First, the outer summation iterates through all possible time points 'a' (i.e., when calculating the autocorrelation contribution between two time points, all possible time points in the complete time series that are 'a'). The inner summation iterates through all possible time points 'b' (i.e., relative to time point 'a', when calculating the autocorrelation contribution between two time points, all possible time points relative to time point 'a' in the complete time series that are 'b'). Alternatively, it can be expressed as the average of the autocorrelation contributions for the time point pair (a, b) (where (a, b) represents the two time points mentioned above, since time point a is earlier than time point b). Then, the outer summation iterates through all possible time points in the entire time series that could be considered part of a time point pair, and the inner summation iterates through all possible time points in the entire time series that could be considered part of a time point pair, calculating their autocorrelation contribution values, and then... Standardization is performed to obtain the average autocorrelation fluctuation coefficient of the i-th frequent itemset at the j-th time scale (i.e., the autocorrelation fluctuation characteristics of the i-th frequent itemset data at consecutive time points at the j-th time scale as described above).

[0120] The inner summation is: ;

[0121] The outer summation is ;

[0122] The aforementioned time point pair (a, b) implies a periodic data fluctuation pattern. In the embodiments of this application, a time decay factor is introduced to give higher weight to fluctuation comparisons at similar time points. For example, for frequent itemset data related to database transaction logs, its autocorrelation fluctuation characteristics may show obvious periodic patterns on a minute-level time scale, indicating the existence of regular batch processing tasks. During calculation, for the complete time series of each frequent itemset data at each time scale, any two different time points are selected for comparative analysis. The autocorrelation contribution value between the a-th time point and the b-th time point is calculated, where time point b must be after time point a. The autocorrelation contribution value is calculated using an exponential decay model, so that point pairs with closer time intervals receive higher weights. Then, all valid time point pairs are traversed to calculate the average autocorrelation fluctuation characteristic as the autocorrelation fluctuation index of the frequent itemset data at that time scale, examining the temporal regularity of data access, capturing the periodic access characteristics of each frequent itemset itself, and identifying the essential difference between regular fluctuations generated by timed batch processing tasks and irregular fluctuations generated by random access.

[0123] Step S423: (Based on 3D feature tensor) At a fixed time point k, calculate the instantaneous coherence of any two frequent itemset data at the same time scale (instantaneous coherence reflects the degree of difference in access fluctuation between the two frequent itemset data at that time point). Average the instantaneous coherence of all time points to obtain the coherent fluctuation characteristics of each pair of frequent itemset data. For each frequent itemset data, use the coherent fluctuation characteristics to calculate its average coherent fluctuation characteristics with all other frequent itemset data.

[0124] It should be noted that, in the above embodiments of this application, at a specific time point k, the instantaneous coherence degree between the first and second frequent itemset data is calculated. This value is the absolute value of the difference between their I / O request frequencies at that time point, divided by (the larger of their I / O request frequencies at that time point, plus a very small positive number to prevent division by zero). Then, the coherence fluctuation characteristic between the first and second frequent itemset data is calculated. This value is the sum of the instantaneous coherence degrees between these two frequent itemset data at all time points, divided by the total number of time points. Finally, for a given frequent itemset data, the average coherence fluctuation characteristic is calculated by first calculating... The co-variance characteristic value between the frequent itemset data and every other frequent itemset data except itself is calculated. Then, all these co-variance characteristic values ​​are summed and finally divided by (the total number of frequent itemset data minus one). In the above embodiment of this application, at each specific time point, the instantaneous co-variance degree between any two different frequent itemset data is calculated. Then, the instantaneous co-variance degree at all time points is averaged to obtain the co-variance characteristic of each pair of frequent itemset data. For each frequent itemset data, its average co-variance characteristic with all other frequent itemset data is calculated. By analyzing the access fluctuation co-variance of different frequent itemset data at the same time, potential group access patterns are discovered.

[0125] Step S424: Calculate the volatility coefficient of frequent itemset data by weighted summation based on the autocorrelation volatility characteristics and the average cooperative volatility characteristics;

[0126] It should be noted that in the above embodiments of this application, the original IO request frequency data is first restructured according to three dimensions: time point, frequent itemset, and data time scale, to establish a three-dimensional feature tensor. This enables subsequent analysis to systematically track and compare access patterns of various frequent itemsets at different time granularities within a unified data framework. Furthermore, by analyzing the relationship between the access frequency changes of the same frequent itemset at different time points, the temporal regularity of data access is examined, capturing the periodic access characteristics of each frequent itemset and identifying the essential difference between regular fluctuations generated by timed batch processing tasks and irregular fluctuations generated by random access. Furthermore, by analyzing the correlation of access behavior of different frequent itemsets at the same time, the collaborative access patterns between multiple frequent itemsets are reflected, and potential group access patterns are identified.

[0127] Finally, the volatility coefficient of frequent itemset data is calculated by weighted summation based on the autocorrelation volatility characteristics and the average co-variance volatility characteristics. The volatility coefficient of frequent itemset data reflects the access stability of frequent itemsets themselves, and also considers their sensitivity to the access volatility of associated data, providing a comprehensive basis for access stability decision-making for data layout optimization.

[0128] In summary, the data block association data storage method proposed in this invention for improving network security availability firstly identifies frequently accessed data block sets by analyzing historical IO requests, forming highly associated data combinations. Then, it comprehensively evaluates the migration value of each frequent itemset by calculating IO latency benefit factors, space cost factors, and collaborative access intensity. Furthermore, it constructs a conflict graph model to analyze resource competition relationships between different frequent itemsets, using interference-aware weights to correct expected utility values ​​and identify and avoid potential resource conflicts. Further, based on the corrected utility values, it employs particle swarm optimization to select the most frequent itemsets with the highest utility values ​​under high-speed storage space constraints, sequentially including them in the migration list to ensure maximum overall benefit with limited resources. Finally, it executes data block migration in batches according to the priority order of the candidate list, improving storage performance while ensuring system stability.

[0129] In the specific execution process, the competitive relationships between frequent itemset data are aggregated, and the propagation depth and range of the competitive impact relationship are combined to analyze the information on the performance degradation of frequent itemset data, and comprehensively analyze the diffusion effect of the fatigue state of fatigued itemset data.

[0130] In addition, by combining the access fluctuation analysis of IO requests, the fluctuation coefficient of frequent itemset data is analyzed, potential group access patterns are identified, and the sensitivity of the data to the fluctuation of access to related data is considered, providing a comprehensive basis for access stability decision-making for data layout optimization.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; those skilled in the art can modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for storing data block associations to improve network security availability, characterized in that, The following steps are included: Multiple IO requests from the server terminal are collected at a preset time period to obtain an IO request data set; the IO request data set is preprocessed to obtain preprocessed IO requests; Obtain the data block combination corresponding to the preprocessed IO request; The data block combinations corresponding to the preprocessed IO requests are processed through association analysis to obtain multiple frequent itemset data, where each frequent itemset data includes a list of data blocks, support, lift, and total itemset capacity. Obtain the data block read / write frequency and storage medium performance parameters corresponding to frequent itemset data; calculate the IO latency benefit factor based on the data block read / write frequency and the storage medium latency difference calculated based on the storage medium performance parameters. The space cost factor is calculated based on the total capacity of the itemset and the preset high-speed storage media allocation threshold. Calculate the collaborative access strength based on support and lift. The expected utility value is calculated by combining the IO latency benefit factor, the space cost factor, and the intensity of collaborative access. Construct a conflict graph model with frequent itemset data as vertices based on all IO requests; and calculate the weights of edges in the conflict graph model based on the competition data among all IO requests. The interference-aware weights for each frequent itemset are calculated based on the conflict graph model. The expected utility value is corrected by using the interference perception weight to obtain the corrected expected utility value; Based on the corrected expected utility values, frequent itemset data are analyzed through particle swarm optimization, and a data migration candidate list is output. Perform data block migration processing based on the data migration candidate list.

2. The data block association data storage method for improving network security availability according to claim 1, characterized in that, Construct a conflict graph model with frequent itemset data as vertices based on all IO requests; and calculate the weights of edges in the conflict graph model based on the competition data among all IO requests. The interference-aware weights for each frequent itemset are calculated based on the conflict graph model. The expected utility value is corrected using interference-perceived weights to obtain the corrected expected utility value, which includes the following steps: Traverse the frequent itemset data involved in each IO request, and determine whether any two frequent itemsets appear in the same IO request or whether the data blocks involved in the two frequent itemsets are accessed simultaneously within the same time window. If any of the above conditions are met, add a competing edge between the corresponding two frequent itemsets. Build a conflict graph model based on all frequent itemsets and competing edges. The weights of each competing edge in the conflict graph model are calculated by weighted summation based on the frequency of occurrence of each pair of frequent itemsets in the same IO request, the frequency of occurrence in the same time window, and the proportion of time occupied by two frequent itemsets simultaneously in the IO request queue. By utilizing the edge weights of the conflict graph model and combining them with fatigue analysis of competition interference, the interference-perceived weight of each frequent itemset data is calculated. The perturbation perception weight represents the degree of perturbation experienced by the frequent itemset in a competitive environment or its ability to perturb other frequent itemsets, and is used to subsequently adjust the expected utility value. The expected utility value is updated based on the interference-aware weights to obtain the corrected expected utility value.

3. A data block association data storage method for improving network security availability according to claim 2, characterized in that, The conflict graph model is represented as: G=(V,E), where V is the set of vertices of frequent itemset data and E is the set of competing edges.

4. A data block association data storage method for improving network security availability according to claim 3, characterized in that, Using the edge weights of the conflict graph model and combining fatigue analysis of competition interference, the interference-perceived weight of each frequent itemset is calculated, including the following steps: For each vertex in the conflict graph model, the sum of the weights of all connecting edges of each vertex is calculated to obtain the basic disturbance intensity. Taking each vertex in the conflict graph model as the center, calculate the sum of the basic interference intensity of all adjacent vertices within the preset propagation depth range to obtain the propagation range intensity. The frequency sequence of each frequent itemset data in each IO request within a continuous time window is obtained. The standard deviation of the frequency sequence is calculated to obtain the time stability index. The fatigue accumulation rate is calculated by weighting and summing the time stability index, propagation range intensity, and basic interference intensity. Extract the baseline performance parameters, safety factor values, and real-time load rate of each frequent itemset data; The dynamic fatigue trigger threshold is calculated based on the baseline performance parameters, safety factor value, and real-time load rate. The fatigue state level of each frequent itemset data is determined by using the dynamic fatigue trigger threshold and fatigue accumulation rate. Fatigue status levels include normal status, warning status, and fatigue status; The target interference perception weight of each frequent itemset data is obtained by analyzing and calculating the fatigue state level, the basic interference intensity, and the edge weight.

5. A data block association data storage method for improving network security availability according to claim 4, characterized in that, The target interference perception weights for each frequent itemset are calculated based on the fatigue state level, the basic interference intensity, and the edge weights. The calculation includes the following steps: The duration of fatigue for each frequent itemset data point during the collection period of an IO request is calculated. Frequent itemset data whose fatigue duration exceeds the fatigue duration threshold are identified as target fatigue frequent itemset data; Acquire the propagation impact frequent itemset data within the preset propagation depth range of the frequent itemset data of fatigue for each target; Obtain the current performance parameters of each frequently affected itemset; calculate the performance degradation ratio of each frequently affected itemset based on the current performance parameters and the baseline performance parameters. The fatigue influence range of each target fatigue frequent itemset is calculated by using the performance degradation ratio and the edge weights between the target fatigue frequent itemset data and each propagation influence frequent itemset data. The fatigue impact range value of each target fatigue frequent itemset data is used as the original interference perception weight. A fixed fatigue state coefficient is set for all frequent itemset data other than the target fatigue frequent itemset data using the ratio of fatigue accumulation rate to dynamic fatigue trigger threshold. The original disturbance perception weights of each frequent itemset data other than the target fatigue frequent itemset data are calculated by multiplying the fixed fatigue state coefficients with the basic disturbance pre-degree of each frequent itemset data. Normalize all the original interference sensing weights to obtain the target interference sensing weights for each frequent itemset data.

6. A data block association data storage method for improving network security availability according to claim 5, characterized in that, Based on the corrected expected utility values, frequent itemset data are analyzed using particle swarm optimization to output a candidate list of data migrations, including the following steps: Initialize the particle swarm parameters, which include the iteration counter and the maximum number of iterations threshold, the initial particle swarm, the initial learning rate, and the initial particle search speed. The iteration counter is initially set to 0; the initial particle swarm consists of N particles, each particle represents a candidate migration list, the vector length of each particle represents the number of frequent itemset data, and the position of each particle vector represents whether the frequent itemset data is migrated or not. Based on the frequency of IO requests to the frequent itemset data corresponding to each particle, an access fluctuation analysis is performed to obtain the fluctuation coefficient of the frequent itemset data. Migration fitness is calculated using the volatility coefficient of frequent itemset data, the corrected expected utility value, and the total itemset capacity of each frequent itemset data. The target data migration candidate list is obtained by iteratively judging and updating the output based on the migration fitness and the iteration counter.

7. A method for storing data block associations to improve network security availability according to claim 6, characterized in that, The target data migration candidate list is obtained by iteratively judging and updating the output based on the migration fitness and the iteration counter, including the following operation steps: Determine if it is the first iteration. If so, take the current migration fitness of each particle as the individual optimal fitness of each particle, and select the migration fitness with the largest migration fitness among all migration fitnesss as the global optimal fitness. If it is not the first iteration, increment the iteration count by 1 to get the current iteration count. Compare the migration fitness of each particle with the individual optimal fitness of each particle at the current iteration count, and select the larger migration fitness as the new individual optimal fitness. Select the largest individual optimal fitness among all the individual optimal fitness of all particles as the new global optimal fitness at the current iteration count. Determine if the current iteration number is greater than or equal to the maximum iteration number threshold. If yes, output the particle corresponding to the new global optimal fitness at the current iteration number as the target data migration candidate list. If no, calculate if the increase ratio of the new global optimal fitness at the current iteration number to the global optimal fitness at the previous iteration number is less than or equal to the increase ratio threshold. If yes, output the particle corresponding to the new global optimal fitness at the current iteration number as the target data migration candidate list. If no, update the initial particle search velocity using the individual optimal fitness at the current iteration number, the global optimal fitness, and the learning rate to obtain a new initial particle search velocity. Update the initial particle swarm using the new particle search velocity to obtain a new initial particle swarm. Return the new initial particle swarm and the new particle search velocity to the above steps for re-iteration until the target data migration candidate list is output.

8. A method for storing data block associations to improve network security availability according to claim 7, characterized in that, Based on the I / O request frequency of the frequent itemset data corresponding to each particle, an access fluctuation analysis is performed to obtain the fluctuation coefficient of the frequent itemset data, including the following steps: The frequency of IO requests for each frequent itemset data at consecutive time points is extracted from the preprocessed IO requests. Based on the frequency of IO requests at consecutive time points, a three-dimensional feature tensor T[k][i][j] is constructed at multiple time scales; where k represents the time point index, i represents the frequent itemset data index, and j represents the time scale index. Each three-dimensional feature tensor T[k][i][j] represents the frequency of IO requests for the i-th frequent itemset data at the k-th time point at the j-th time scale. Based on the three-dimensional feature tensor, calculate the autocorrelation fluctuation characteristics of the i-th frequent itemset data at consecutive time points under the j-th time scale; At a fixed time point k, calculate the instantaneous coherence of any two frequent itemsets at the same time scale. Average the instantaneous coherence of all time points to obtain the coherent fluctuation characteristics of each pair of frequent itemsets. For each frequent itemset, use the coherent fluctuation characteristics to calculate its average coherent fluctuation characteristics with all other frequent itemsets. The volatility coefficient of frequent itemset data is calculated by weighted summation based on the autocorrelation volatility characteristics and the average co-variance volatility characteristics.

9. A method for storing data block associations to improve network security availability according to claim 8, characterized in that, The calculation method of the autocorrelation fluctuation feature is as follows: for each frequent itemset in the complete time series at each time scale, by selecting two different time points for comparative analysis, the autocorrelation contribution value between the two time points is obtained. Then, by traversing the autocorrelation contribution values ​​of every two time points in the complete time series, the autocorrelation fluctuation feature is obtained.

10. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steps of the data block association data storage method for improving network security availability as described in any one of claims 1-9.

Images