Bitmap index container optimization method and system based on multi-dimensional dynamic decision

Abstract

The application provides a bitmap index container optimization method and system based on multi-dimensional dynamic decision, relates to the technical field of databases, and comprises the following steps: a skewness coefficient value is calculated, a feature fusion vector is constructed by fusing a data distribution feature vector, a data access rule vector and a change trend vector, a double strategy gradient algorithm is used for iterative optimization to obtain an adaptive threshold value, it is judged whether to switch the container type, data deduplication, priority sorting are performed through a temporary container, batch migration data is performed, and a bidirectional index table is established to redirect an access request.

Description

Technical Field

[0001] This invention relates to the field of database technology, and in particular to a bitmap index container optimization method and system based on multidimensional dynamic decision-making. Background Technology

[0002] With the rapid development of big data applications, bitmap index technology in database systems is widely used to improve query efficiency and data processing performance. Bitmap index containers, as important carriers for storing and managing bitmap indexes, have a significant impact on overall system performance.

[0003] In traditional database applications, bitmap indexes are mainly used for fast retrieval of low cardinality data columns. With the expansion of application scenarios, bitmap indexes have been widely used in various environments. Traditional bitmap index container designs usually adopt static and fixed container types, which cannot be dynamically adjusted according to changes in data characteristics and access patterns.

[0004] The distribution characteristics, access frequency, and trends of data can change significantly over time and in different business scenarios. Fixed container types are difficult to adapt to dynamically changing needs, especially in high-concurrency, large-scale data processing scenarios. Inappropriate container type selection can lead to a significant decrease in system performance. Summary of the Invention

[0005] This invention provides a bitmap index container optimization method and system based on multidimensional dynamic decision-making, which can at least solve some of the problems existing in the prior art.

[0006] A first aspect of this invention provides a bitmap index container optimization method based on multidimensional dynamic decision-making, comprising:

[0007] Obtain the data feature values ​​of the bitmap index container and calculate the skewness coefficient value based on the data feature values;

[0008] Data distribution feature vectors are obtained by processing data feature values ​​through a pre-set time series diagram structure. Data access pattern vectors and corresponding data change trend vectors are determined based on the data feature values. Data distribution feature vectors, data access pattern vectors, and data change trend vectors are combined across dimensions to obtain feature fusion vectors. Based on the feature fusion vectors, decision parameters are iteratively optimized using a dual-strategy gradient algorithm to obtain the adaptive threshold for container switching.

[0009] The skewness coefficient value is compared with the adaptive threshold to determine whether to switch the container type.

[0010] When container type switching is required, a temporary container is created to perform deduplication on the data to be migrated in the bitmap index container to obtain deduplicated data. The popularity value and relevance value of the deduplicated data are calculated. The weighted sum of the popularity value and relevance value is used as the priority value and the deduplicated data is sorted to obtain the sequence of data to be migrated. The sequence of data to be migrated is written to the temporary container in batches and the write status value is recorded. When the write status value indicates that the pre-commit is successful, the confirmation commit is executed, a bidirectional index table is generated, and the bidirectional index table is used to redirect data access requests.

[0011] Monitor the performance metrics after the switch, and when the performance metrics exceed the preset range, perform a rollback operation and update the decision parameters.

[0012] In one alternative implementation,

[0013] Obtaining the data feature values ​​of a bitmap indexed container and calculating the skewness coefficient value based on the data feature values ​​includes:

[0014] Within a preset first time window, the number of writes and reads of bitmap cells in the bitmap index container is counted, the distribution density of data storage locations in the bitmap cells is calculated, and data feature values ​​are obtained.

[0015] The data feature values ​​are divided into multiple data segments according to preset rules, and the distribution characteristics of each data segment are calculated. The segments are then combined to obtain a probability distribution sequence.

[0016] The skewness coefficient value corresponding to the bitmap index container is calculated based on the probability distribution sequence.

[0017] In one alternative implementation,

[0018] Data distribution feature vectors are obtained by processing data feature values ​​through a pre-set time series graph structure. Based on the data feature values, data access pattern vectors and corresponding data change trend vectors are determined. The feature fusion vector is obtained by combining the data distribution feature vector, data access pattern vector, and data change trend vector across dimensions.

[0019] A time series graph structure is constructed based on the data feature values ​​based on a pre-set time interval and multiple time nodes are recorded. The bitmap cell state parameters of each time node are calculated. A state transition matrix is ​​constructed based on the bitmap cell state parameters and feature decomposition is performed to obtain the data distribution feature vector. The bitmap cell state parameters include occupancy status, modification time and index information.

[0020] Data blocks in bitmap units are obtained from the time sequence diagram structure. The standard access count of the data blocks within a preset time period and the consecutive access count between the data blocks are recorded. The ratio of the consecutive access count to the standard access count is used to construct an access pattern matrix. The standardized Laplace matrix of the access pattern matrix is ​​calculated and the feature vector is extracted as a data access pattern vector.

[0021] The time window sequence is divided according to a preset time length. The number of data operations in each time window is recorded. The ratio of the number of data operations to the total number of operations in the time window is calculated to obtain the operation ratio parameter. The operation ratio parameter is multiplied by the corresponding preset weight coefficient and summed to obtain the trend index. The trend index is subjected to time series analysis to obtain the data trend vector. The number of data operations includes the number of data writes, the number of data updates, and the number of data deletions.

[0022] The data distribution feature vector, data access pattern vector, and data change trend vector are dimensionally aligned and normalized, and the results are weighted and summed to obtain the feature fusion vector.

[0023] In one alternative implementation,

[0024] Based on the aforementioned feature fusion vector, and by iteratively optimizing the decision parameters using a dual-policy gradient algorithm, the adaptive threshold for container switching is obtained, including:

[0025] Historical data is acquired to construct a historical data sequence. Based on the length of the historical data sequence, a time decay coefficient sequence is constructed as a time decay weight. The gradient of each feature in the feature fusion vector with respect to the pre-set initial policy function is calculated. The L2 norm of the gradient is multiplied by the time decay weight and accumulated to obtain the importance score of each feature.

[0026] Based on the importance score, an attention transformation matrix group is constructed. The feature fusion vector is multiplied by the attention transformation matrix group to obtain the query vector, key vector, and numerical vector. The dot product of the query vector and the key vector is calculated, and after softmax normalization, it is multiplied by the numerical vector to obtain the reconstructed feature representation. One-dimensional convolution operation and ReLU function activation are performed on the reconstructed feature representation. The compressed feature representation is obtained by mapping through a fully connected layer. The feature selection mask is determined by the sigmoid function and multiplied element-wise with the compressed feature representation to obtain the selected feature representation.

[0027] A random noise vector is constructed based on a Gaussian distribution. The selected feature representation is connected to the random noise vector to form a feature supplement input. A feature supplement value is generated by combining nonlinear transformation. The feature supplement value is added to the selected feature representation to obtain the supplemented feature representation.

[0028] The policy gradient corresponding to the supplemented feature representation is calculated based on the initial policy function. The policy gradient is multiplied by the pre-set adaptive learning rate to update the parameters in the initial policy function, and the adaptive threshold for container switching is calculated based on the updated policy parameters.

[0029] In one alternative implementation,

[0030] When container type switching is required, a temporary container is created to perform deduplication on the data to be migrated in the bitmap index container. The deduplicated data is then calculated based on its popularity and relevance. The weighted sum of these two values ​​is used as a priority value, and the deduplicated data is sorted to obtain the sequence of data to be migrated, which includes:

[0031] When container type switching is required, a temporary container is created and the data to be migrated from the bitmap index container is input into the temporary container. The feature fingerprint of each data in the data to be migrated is calculated using the locality sensitive hashing algorithm. Based on the feature fingerprint, the data similarity between any two data in the data to be migrated is calculated. The data similarity is compared with a preset similarity threshold. When the data similarity is greater than the preset similarity threshold, it is determined to be duplicate data and removed to obtain deduplicated data.

[0032] The access frequency is obtained by counting the number of times the deduplicated data is accessed within the second sliding time window. The time weight is obtained by calculating the time interval between the last access and the current time. The survival time is obtained by calculating the time interval between the data generation time and the current time. The popularity value is calculated based on the access frequency, the time weight and the survival time.

[0033] The association weight is obtained by calculating the reference relationship between each pair of data in the deduplicated data, the semantic similarity between each pair of data is calculated, and the association weight and the semantic similarity are multiplied and the average value is calculated to obtain the association degree value.

[0034] According to the task requirements, the weight coefficients corresponding to the popularity value and the relevance value are initialized and weighted summed to obtain the priority value. Based on the priority value, the deduplicated data is sorted in descending order to obtain the data sequence to be migrated.

[0035] In one alternative implementation,

[0036] The data sequence to be migrated is written to a temporary container in batches, and the write status value is recorded. When the write status value indicates that the pre-commit was successful, a confirmation commit is executed, a bidirectional index table is generated, and the bidirectional index table is used to redirect data access requests, including:

[0037] The data sequence to be migrated is grouped into multiple data batches according to a preset batch size. The multiple data batches are written into a temporary container, and a status value is recorded for each data batch.

[0038] Calculate the hash value of the data in each data batch and sum them to obtain the data checksum. Compare the data checksum after writing with the data checksum before writing to obtain the integrity verification result. Generate a pre-commit log record based on the integrity verification result. When the write status value indicates that the pre-commit is successful, execute the confirmation commit operation.

[0039] Obtain the original data identifier of the data to be migrated from the bitmap index container, obtain the new data identifier of the data to be migrated from the temporary container, establish a mapping relationship and add version number and timestamp to generate metadata information, and construct a bidirectional index table based on the original data identifier, the new data identifier and the metadata information;

[0040] The mapping relationship between the original data identifier and the new data identifier is stored in the redirection cache. When a data access request is received, the target data location is first searched in the redirection cache. If the target data location is not found, the target data location is searched in the bidirectional index table. If the search fails, it is determined whether the number of retries exceeds the search threshold. If it does not exceed the threshold, a retry is performed. If it exceeds the threshold and the operation is a read operation, the data in the bitmap index container is accessed. If the threshold is exceeded and the operation is a write operation, the data access request is rejected.

[0041] In one alternative implementation,

[0042] Monitor performance metrics after the switchover. When a performance metric exceeds a preset range, perform a rollback operation and update decision parameters, including:

[0043] Obtain the performance metrics after the switch, including service response time, resource utilization, and service success rate;

[0044] Determine whether the performance index value exceeds a preset performance threshold range. The preset performance threshold range is determined based on the statistical distribution of historical performance data. When the duration for which the performance index value continuously exceeds the preset performance threshold range is greater than a preset time threshold, a rollback operation is triggered.

[0045] The rollback operation is executed and the decision parameters are updated based on the performance fluctuations of this switch. The decision parameters include the switch time window, performance monitoring frequency, and performance threshold range.

[0046] A second aspect of the present invention provides a bitmap index container optimization system based on multidimensional dynamic decision-making, comprising:

[0047] The first unit is used to obtain the data feature values ​​of the bitmap index container and calculate the skewness coefficient value based on the data feature values;

[0048] The second unit is used to process data feature values ​​through a pre-set time series diagram structure to obtain a data distribution feature vector, determine a data access pattern vector and a corresponding data change trend vector based on the data feature values, combine the data distribution feature vector, data access pattern vector and data change trend vector across dimensions to obtain a feature fusion vector, and iteratively optimize the decision parameters based on the feature fusion vector and a dual-strategy gradient algorithm to obtain an adaptive threshold for container switching.

[0049] The third unit is used to compare the skewness coefficient value with the adaptive threshold and determine whether to switch the container type.

[0050] The fourth unit is used to create a temporary container to perform deduplication on the data to be migrated in the bitmap index container when container type switching is required. It calculates the popularity and relevance values ​​of the deduplicated data, uses the weighted sum of the popularity and relevance values ​​as the priority value, sorts the deduplicated data to obtain the sequence of data to be migrated, writes the sequence of data to be migrated into the temporary container in batches and records the write status value. When the write status value indicates that the pre-commit is successful, it performs a confirmation commit, generates a bidirectional index table, and uses the bidirectional index table to redirect data access requests.

[0051] The fifth unit is used to monitor the performance index values ​​after the switch. When the performance index values ​​exceed the preset range, a rollback operation is performed and the decision parameters are updated.

[0052] A third aspect of the present invention provides an electronic device, comprising:

[0053] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0054] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0055] In this invention, by acquiring the data feature values ​​of the bitmap index container and calculating the skewness coefficient, and combining them with the multidimensional feature vector obtained through time-series graph structure processing, a dual-strategy gradient algorithm is used to optimize the decision parameters, thereby realizing intelligent decision-making for container type switching. This effectively improves the system's adaptability to different data distribution scenarios. By creating temporary containers to perform deduplication operations and calculating data priority based on heat value and correlation value, orderly data migration is achieved. At the same time, a bidirectional index table is introduced to redirect access requests, ensuring the smoothness of the container switching process and the continuity of data access, reducing the risk of system switching. A performance monitoring and rollback mechanism is introduced, automatically executing a rollback operation when the performance index value after switching exceeds the preset range, and forming a closed-loop optimization by updating the decision parameters, thereby improving the stability and reliability of the system. Meanwhile, by continuously learning and optimizing the decision model, the system's adaptive capability is enhanced. Attached Figure Description

[0056] Figure 1 This is a flowchart illustrating the bitmap index container optimization method based on multidimensional dynamic decision-making according to an embodiment of the present invention.

[0057] Figure 2 This is a data migration logic flowchart of the bitmap index container optimization method based on multi-dimensional dynamic decision-making in an embodiment of the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0059] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0060] Figure 1 This is a flowchart illustrating the bitmap index container optimization method based on multi-dimensional dynamic decision-making according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0061] Obtain the data feature values ​​of the bitmap index container and calculate the skewness coefficient value based on the data feature values;

[0062] Data distribution feature vectors are obtained by processing data feature values ​​through a pre-set time series diagram structure. Data access pattern vectors and corresponding data change trend vectors are determined based on the data feature values. Data distribution feature vectors, data access pattern vectors, and data change trend vectors are combined across dimensions to obtain feature fusion vectors. Based on the feature fusion vectors, decision parameters are iteratively optimized using a dual-strategy gradient algorithm to obtain the adaptive threshold for container switching.

[0063] The skewness coefficient value is compared with the adaptive threshold to determine whether to switch the container type.

[0064] When container type switching is required, a temporary container is created to perform deduplication on the data to be migrated in the bitmap index container to obtain deduplicated data. The popularity value and relevance value of the deduplicated data are calculated. The weighted sum of the popularity value and relevance value is used as the priority value and the deduplicated data is sorted to obtain the sequence of data to be migrated. The sequence of data to be migrated is written to the temporary container in batches and the write status value is recorded. When the write status value indicates that the pre-commit is successful, the confirmation commit is executed, a bidirectional index table is generated, and the bidirectional index table is used to redirect data access requests.

[0065] Monitor the performance metrics after the switch, and when the performance metrics exceed the preset range, perform a rollback operation and update the decision parameters.

[0066] In one alternative implementation,

[0067] Obtaining the data feature values ​​of a bitmap indexed container and calculating the skewness coefficient value based on the data feature values ​​includes:

[0068] Within a preset first time window, the number of writes and reads of bitmap cells in the bitmap index container is counted, the distribution density of data storage locations in the bitmap cells is calculated, and data feature values ​​are obtained.

[0069] The data feature values ​​are divided into multiple data segments according to preset rules, and the distribution characteristics of each data segment are calculated. The segments are then combined to obtain a probability distribution sequence.

[0070] The skewness coefficient value corresponding to the bitmap index container is calculated based on the probability distribution sequence.

[0071] The read / write characteristics of bitmap cells in a bitmap index container are analyzed. Within a preset first time window, the number of writes and reads for each bitmap cell is recorded. The time window can be set to 30 minutes. During this period, the monitoring module statistically analyzes the operations of each bitmap cell in the bitmap index container. For example, in a bitmap index container containing 1000 bitmap cells, within a 30-minute time window, bitmap cell 1 is recorded as having 150 writes and 300 reads; bitmap cell 2 is recorded as having 80 writes and 450 reads, and so on.

[0072] The distribution density of data storage locations within a bitmap cell is calculated. Distribution density represents the distribution of data across different storage locations within the bitmap cell. The calculation method involves counting the frequency of data occurrence at each storage location and dividing by the total number of storage locations. For example, in a bitmap cell containing 256 storage locations, assume the data occurrence frequency is 0.4 at locations 1-50, 0.3 at locations 51-150, and 0.2 at locations 151-256. By analyzing the statistical data, the data characteristic values ​​of this bitmap index container can be obtained, including write frequency, read frequency, and data distribution density.

[0073] The acquired data feature values ​​are divided into multiple data segments according to preset rules. The division of data segments can be based on the distribution range of the data feature values. For example, the write frequency can be divided into a low-frequency segment (0-50 times / 30 minutes), a medium-frequency segment (51-200 times / 30 minutes), and a high-frequency segment (more than 201 times / 30 minutes); the read frequency can be divided into a low-frequency segment (0-100 times / 30 minutes), a medium-frequency segment (101-500 times / 30 minutes), and a high-frequency segment (more than 501 times / 30 minutes); and the data distribution density can be divided into a low-density segment (0-0.3), a medium-density segment (0.31-0.7), and a high-density segment (0.71-1.0).

[0074] For each data segment, calculate the distribution characteristics. The distribution characteristics include the proportion of bitmap cells in that data segment to the total number of bitmap cells. For example, in the low-frequency segment (0-50 writes / 30 minutes), there are 400 bitmap cells, so the distribution characteristics of this data segment are 400 / 1000 = 0.4; in the mid-frequency segment (51-200 writes / 30 minutes), there are 500 bitmap cells, so the distribution characteristics of this data segment are 500 / 1000 = 0.5; and in the high-frequency segment (above 201 writes / 30 minutes), there are 100 bitmap cells, so the distribution characteristics of this data segment are 100 / 1000 = 0.1.

[0075] The distribution characteristics of each data segment are combined to form a probability distribution sequence. Taking the write frequency as an example, the probability distribution sequence is [0.4, 0.5, 0.1], representing the distribution characteristics of the low-frequency, mid-frequency, and high-frequency segments. Similarly, the probability distribution sequence for the read frequency may be [0.3, 0.6, 0.1], and the probability distribution sequence for the data distribution density may be [0.2, 0.5, 0.3].

[0076] The skewness coefficient of a bitmap index container is calculated based on a probability distribution sequence. The skewness coefficient measures the symmetry of the data distribution. It is calculated by dividing the sum of the cubes of the differences between each element in the probability distribution sequence and the median by the cube of the standard deviation, and then multiplying by a scaling factor based on the number of data segments. For a probability distribution sequence of write frequencies [0.4, 0.5, 0.1], with a median of 0.4, a standard deviation of approximately 0.2, and 3 data segments, the skewness coefficient is ((0.4-0.4)^3+(0.5-0.4)^3+(0.1-0.4)^3) / (0.2^3)*(3 / (3-1))=-0.5. Negative skewness indicates that the data distribution is biased towards higher values.

[0077] For example, suppose there is a bitmap index container containing 10,000 bitmap cells. Within a 30-minute time window, the write frequency distribution of the bitmap cells is as follows: 4,000 cells have 0-50 writes, 5,000 cells have 51-200 writes, and 1,000 cells have more than 201 writes; the read frequency distribution is as follows: 2,000 cells have 0-100 reads, 7,000 cells have 101-500 reads, and 1,000 cells have more than 501 reads; the data density distribution is as follows: 3,000 cells have a density of 0-0.3, 5,000 cells have a density of 0.3-0.7, and 2,000 cells have a density of 0.7-1.0.

[0078] Based on this, the probability distribution sequence of the write frequency is calculated to be [0.4, 0.5, 0.1], the probability distribution sequence of the read frequency is [0.2, 0.7, 0.1], and the probability distribution sequence of the data distribution density is [0.3, 0.5, 0.2].

[0079] The calculated skewness coefficients for write frequency, read frequency, and data distribution density are -0.5, -0.3, and -0.15, respectively. These skewness coefficient values ​​indicate that the write frequency, read frequency, and data distribution density of the bitmap cells in the bitmap index container all exhibit a slightly negative skewed distribution, meaning that the data is slightly concentrated in the high-value region.

[0080] In this embodiment, by real-time statistical analysis of the read and write counts of bitmap units and combined with the distribution density analysis of data storage locations, the data access patterns and hotspot distribution can be accurately grasped, and data access imbalances can be detected in a timely manner, providing a basis for resource scheduling optimization. The analysis method combining data segmentation and probability distribution sequences can not only meticulously depict the local characteristics of data distribution, but also intuitively reflect the overall data skewness through the skewness coefficient, providing a reliable decision-making basis for system storage optimization and query optimization. Data segmentation through preset rules makes the analysis process standardized and simplifies engineering implementation.

[0081] In one alternative implementation,

[0082] Data distribution feature vectors are obtained by processing data feature values ​​through a pre-set time series graph structure. Based on the data feature values, data access pattern vectors and corresponding data change trend vectors are determined. The feature fusion vector is obtained by combining the data distribution feature vector, data access pattern vector, and data change trend vector across dimensions.

[0083] A time series graph structure is constructed based on the data feature values ​​based on a pre-set time interval and multiple time nodes are recorded. The bitmap cell state parameters of each time node are calculated. A state transition matrix is ​​constructed based on the bitmap cell state parameters and feature decomposition is performed to obtain the data distribution feature vector. The bitmap cell state parameters include occupancy status, modification time and index information.

[0084] Data blocks in bitmap units are obtained from the time sequence diagram structure. The standard access count of the data blocks within a preset time period and the consecutive access count between the data blocks are recorded. The ratio of the consecutive access count to the standard access count is used to construct an access pattern matrix. The standardized Laplace matrix of the access pattern matrix is ​​calculated and the feature vector is extracted as a data access pattern vector.

[0085] The time window sequence is divided according to a preset time length. The number of data operations in each time window is recorded. The ratio of the number of data operations to the total number of operations in the time window is calculated to obtain the operation ratio parameter. The operation ratio parameter is multiplied by the corresponding preset weight coefficient and summed to obtain the trend index. The trend index is subjected to time series analysis to obtain the data trend vector. The number of data operations includes the number of data writes, the number of data updates, and the number of data deletions.

[0086] The data distribution feature vector, data access pattern vector, and data change trend vector are dimensionally aligned and normalized, and the results are weighted and summed to obtain the feature fusion vector.

[0087] A time-series graph structure is constructed based on data feature values ​​at pre-set time intervals. The time interval can be set to 5 minutes. Within this interval, data feature values ​​of the bitmap index container are collected, and multiple time nodes are recorded. For example, within a one-hour monitoring period, data feature values ​​for 12 time nodes can be obtained. At each time node, the status parameters of the bitmap cell are calculated, including occupancy status, modification time, and index information. Occupancy status indicates whether the bitmap cell is occupied by data; 0 represents idle, and 1 represents occupied. The modification time records the timestamp of the bitmap cell's most recent modification. The index information contains pointers or references to the actual data. For an index container containing 1000 bitmap cells, at time T1, approximately 700 cells may be in an occupied state, with an average modification time of 10 minutes prior, and the index information pointing to different tables and records in the database.

[0088] A state transition matrix is ​​constructed based on the state parameters of the bitmap cells. The state transition matrix reflects the changes in the state of the bitmap cells between adjacent time points. The rows and columns of the matrix represent different states, and the matrix elements represent the probability of transitioning from one state to another. Assume that the combination of occupied state and modification time is defined as four states: Newly Created (occupied and modification time less than 5 minutes), Active (occupied and modification time between 5 and 30 minutes), Static (occupied and modification time greater than 30 minutes), and Idle (not occupied). In the constructed 4x4 state transition matrix, each element represents the probability of transitioning from one state to another. For example, the matrix element (1, 2) with a value of 0.3 indicates that the probability of transitioning from the "Newly Created" state to the "Active" state is 0.3.

[0089] Eigenvalue decomposition (EVD) of the state transition matrix yields the data distribution eigenvectors. EVD reveals the main patterns of state transitions. EVD of the aforementioned 4x4 state transition matrix yields four eigenvalues ​​and their corresponding eigenvectors. The eigenvector corresponding to the largest eigenvalue is selected as the data distribution eigenvector, for example, [0.2, 0.5, 0.2, 0.1], representing the long-term distribution proportion of bitmap cells in the four states of newly created, active, static, and idle.

[0090] Data blocks are retrieved from bitmap cells in the sequence diagram structure, and their access patterns are recorded within a preset time period. The preset time period can be set to 30 minutes. During this period, the standard access count for each data block and the number of consecutive accesses between data blocks are recorded. The standard access count refers to the total number of times a single data block is accessed; the consecutive access count refers to the number of times one data block is accessed immediately after being accessed from another. For example, data block A was accessed 100 times in 30 minutes, data block B was accessed 80 times, there were 40 consecutive accesses from A to B, and 30 consecutive accesses from B to A.

[0091] The ratio of consecutive accesses to standard accesses is used to construct an access pattern matrix. In the previous example, the ratio from A to B is 40 / 100 = 0.4, and the ratio from B to A is 30 / 80 = 0.375. These ratios form the elements of the access pattern matrix, reflecting the strength of access associations between data blocks.

[0092] The normalized Laplacian matrix of the access pattern matrix is ​​calculated, and eigenvectors are extracted. The normalized Laplacian matrix reveals the relative correlation structure between data blocks. For n data blocks, an n×n normalized Laplacian matrix is ​​constructed by calculating the degree matrix and adjacency matrix according to the definition of the normalized Laplacian matrix. Eigenvalue decomposition is performed on the calculated normalized Laplacian matrix, and the eigenvectors corresponding to the k smallest non-zero eigenvalues ​​are selected and combined to form a data access pattern vector. For example, for 5 data blocks, 2 eigenvectors are selected, resulting in a 5×2 matrix as the data access pattern vector.

[0093] The system divides the data into time windows based on a preset time length and records the number of data operations within each window. The time length can be set to 10 minutes, and the monitoring period can be divided into multiple time windows. Within each time window, the system records the number of data writes, updates, and deletions. For example, in the first 10-minute window, 200 writes, 300 updates, and 50 deletions were recorded, for a total of 550 operations.

[0094] The operation ratio parameter is obtained by calculating the ratio of the number of data operations to the total number of operations within the time window. For the example above, the write operation ratio is 200 / 550=0.364, the update operation ratio is 300 / 550=0.545, and the delete operation ratio is 50 / 550=0.091.

[0095] The trend indicator is obtained by multiplying the operation ratio parameter by the corresponding preset weight coefficient and summing the results. The preset weight coefficient can be set according to the importance of the operation type. For example, the weight of a write operation is 0.5, the weight of an update operation is 0.3, and the weight of a delete operation is 0.2. In the example above, the trend indicator is 0.364×0.5+0.545×0.3+0.091×0.2=0.364.

[0096] Time series analysis of trend indicators yields a data trend vector. This analysis can employ the sliding window method, calculating the differences and rates of change of trend indicators between adjacent time windows. For example, if the trend indicators for six consecutive time windows are 0.364, 0.382, 0.395, 0.410, 0.425, and 0.435, five differences [0.018, 0.013, 0.015, 0.015, 0.01] and rates of change [4.9%, 3.4%, 3.8%, 3.7%, 2.4%] can be calculated. These differences and rates of change are then combined to form the data trend vector.

[0097] The data distribution feature vector, data access pattern vector, and data change trend vector are subjected to dimension alignment and normalization. Dimension alignment can be achieved through principal component analysis or dimensionality reduction techniques to ensure that the three vectors have the same dimension. Normalization can be performed using the minimum-maximum normalization method, mapping the elements of each vector to the interval [0, 1]. For example, the data distribution feature vector [0.2, 0.5, 0.2, 0.1] remains normalized to [0.2, 0.5, 0.2, 0.1], a column in the data access pattern vector [0.1, 0.3, 0.5, 0.2] becomes [0.2, 0.6, 1.0, 0.4] after normalization, and the data change trend vector [0.018, 0.013, 0.015, 0.015] becomes [1.0, 0.722, 0.833, 0.833] after normalization.

[0098] The processed results are weighted and summed to obtain the feature fusion vector. The weights of the three vectors can be set to 0.4, 0.4 and 0.2 respectively. The corresponding elements are multiplied and then summed to obtain the feature fusion vector.

[0099] In this embodiment, by performing temporal analysis and matrix decomposition on the state parameters of bitmap units, the dynamic changes in data distribution can be captured. Constructing a state transition matrix and performing feature decomposition not only reflects the instantaneous state of data distribution but also reveals its evolutionary patterns, providing a more comprehensive basis for storage optimization. Combined with the extraction of eigenvectors from the standardized Laplace matrix, the intrinsic patterns of data access can be deeply explored, effectively guiding the optimization of caching strategies and the improvement of data prefetching mechanisms. The combination of time window sequences and operation ratio parameters enables refined quantification of data operation behavior, providing significant guidance for resource scheduling and load balancing. Through dimension alignment and normalization, feature vectors from multiple dimensions are fused into a unified feature representation, preserving important information from each dimension while achieving effective feature integration, significantly improving the overall performance and stability of the system.

[0100] In one alternative implementation,

[0101] Based on the aforementioned feature fusion vector, and by iteratively optimizing the decision parameters using a dual-policy gradient algorithm, the adaptive threshold for container switching is obtained, including:

[0102] Historical data is acquired to construct a historical data sequence. Based on the length of the historical data sequence, a time decay coefficient sequence is constructed as a time decay weight. The gradient of each feature in the feature fusion vector with respect to the pre-set initial policy function is calculated. The L2 norm of the gradient is multiplied by the time decay weight and accumulated to obtain the importance score of each feature.

[0103] Based on the importance score, an attention transformation matrix group is constructed. The feature fusion vector is multiplied by the attention transformation matrix group to obtain the query vector, key vector, and numerical vector. The dot product of the query vector and the key vector is calculated, and after softmax normalization, it is multiplied by the numerical vector to obtain the reconstructed feature representation. One-dimensional convolution operation and ReLU function activation are performed on the reconstructed feature representation. The compressed feature representation is obtained by mapping through a fully connected layer. The feature selection mask is determined by the sigmoid function and multiplied element-wise with the compressed feature representation to obtain the selected feature representation.

[0104] A random noise vector is constructed based on a Gaussian distribution. The selected feature representation is connected to the random noise vector to form a feature supplement input. A feature supplement value is generated by combining nonlinear transformation. The feature supplement value is added to the selected feature representation to obtain the supplemented feature representation.

[0105] The policy gradient corresponding to the supplemented feature representation is calculated based on the initial policy function. The policy gradient is multiplied by the pre-set adaptive learning rate to update the parameters in the initial policy function, and the adaptive threshold for container switching is calculated based on the updated policy parameters.

[0106] Historical data is acquired to construct a historical data sequence. This historical data includes operation records and performance metrics of the bitmap index container over a past period. For example, collecting data from the last 30 days, sampling 24 times per day, creates a historical data sequence of length 720. A time decay coefficient sequence is then constructed based on the length of this historical data sequence as the time-series decay weights. The time decay coefficient reflects the timeliness of the historical data, with more recent data having higher weights. For a historical data sequence of length 720, an exponential decay method can be used to construct the time-series decay weights, with the most recent data point having a weight of 1.0, decreasing sequentially backward, and the earliest data point having a weight of approximately 0.01.

[0107] The gradient of each feature in the feature fusion vector with respect to a pre-defined initial policy function is calculated. This initial policy function can be a multilayer perceptron network, taking the feature fusion vector as input and outputting the operation decision of a bitmap index container. Assuming the feature fusion vector has a dimension of 10, the initial policy function consists of an input layer (10 nodes), a hidden layer (20 nodes), and an output layer (5 nodes). The gradient of each feature with respect to the policy function output is calculated using the backpropagation algorithm. For example, for the feature fusion vector [0.3, 0.7, 0.2, 0.5, 0.8, 0.4, 0.6, 0.1, 0.9, 0.5], the calculated gradient vector might be [0.05, 0.12, 0.03, 0.08, 0.15, 0.07, 0.09, 0.02, 0.18, 0.08].

[0108] The importance score for each feature is obtained by multiplying the gradient's L2 norm by the temporal decay weights and summing the results. The L2 norm represents the length of the gradient vector and is calculated as the square root of the sum of the squares of the gradient's elements. For the gradient vector above, its L2 norm is approximately 0.31. Multiplying the L2 norm value by the corresponding temporal decay weight (e.g., 0.9) yields 0.28. This calculation is repeated for all times in the historical data sequence, and the results are summed to obtain the importance score for each feature. For example, the importance scores for 10 features might be [12.5, 28.3, 8.7, 19.2, 35.6, 16.8, 21.4, 5.3, 41.2, 18.9].

[0109] An attention transformation matrix set is constructed based on the importance score. The attention transformation matrix set contains three matrices: a query transformation matrix, a key-value transformation matrix, and a numerical transformation matrix. The elements of the matrices are proportionally allocated according to the feature importance score. For example, for a feature fusion vector with dimension 10, three 10×8 transformation matrices are constructed, with features having higher importance scores receiving correspondingly higher weights in the matrices.

[0110] Multiplying the feature fusion vector with the attention transformation matrix yields the query vector, key vector, and numerical vector. Assuming the feature fusion vector is [0.3, 0.7, 0.2, 0.5, 0.8, 0.4, 0.6, 0.1, 0.9, 0.5], multiplying it with the corresponding transformation matrix results in three 8-dimensional vectors: query vector [0.62, 0.54, 0.48, 0.71, 0.39, 0.56, 0.67, 0.43], key vector [0.58, 0.49, 0.52, 0.65, 0.44, 0.51, 0.62, 0.47], and numerical vector [0.60, 0.52, 0.50, 0.68, 0.41, 0.54, 0.64, 0.45].

[0111] The dot product of the query vector and the key vector is calculated, normalized using softmax, and then multiplied by the numerical vector to obtain the reconstructed feature representation. The dot product operation calculates the product of corresponding elements of the query vector and the key vector and sums them to obtain a scalar value. This process is repeated for key vectors at different positions to obtain a set of attention scores. These scores are normalized to a probability distribution using the softmax function, multiplied by corresponding elements of the numerical vector, and summed to obtain the reconstructed feature representation, which is [0.59, 0.53, 0.49, 0.69, 0.42, 0.53, 0.65, 0.44].

[0112] One-dimensional convolution and ReLU activation are performed on the reconstructed feature representation. The one-dimensional convolution operation uses three convolution kernels, each with a width of 3 and a stride of 1, to obtain a 6-dimensional output. The ReLU function sets negative values ​​to 0 and keeps positive values ​​unchanged. For example, the feature representation after convolution and ReLU activation is [1.23, 0.98, 0.87, 1.45, 0.76, 1.02].

[0113] The compressed feature representation is obtained through mapping using a fully connected layer. The fully connected layer maps the 6-dimensional input to a 4-dimensional output, with a parameter matrix of size 6×4. After mapping by the fully connected layer, the compressed feature representation is obtained, for example, [0.85, 1.32, 0.74, 1.17].

[0114] The feature selection mask is determined by the sigmoid function and then multiplied element-wise with the compressed feature representation to obtain the selected feature representation. The sigmoid function maps the values ​​to the range of 0 to 1, which serves as the feature selection mask. For the compressed feature representation, the feature selection mask calculated using the additional weight matrix may be [0.92, 0.78, 0.65, 0.88]. Multiplying the mask element-wise with the compressed feature representation yields the selected feature representation [0.78, 1.03, 0.48, 1.03].

[0115] A random noise vector is constructed based on a Gaussian distribution. The selected feature representation is then concatenated with the random noise vector to form a supplementary feature input. The elements of the random noise vector are sampled from a Gaussian distribution with a mean of 0 and a standard deviation of 0.1, and the dimension is the same as the selected feature representation, for example [0.05, -0.08, 0.12, -0.03]. The selected feature representation is then concatenated with the random noise vector to form an 8-dimensional supplementary feature input [0.78, 1.03, 0.48, 1.03, 0.05, -0.08, 0.12, -0.03].

[0116] Feature supplement values ​​are generated by combining nonlinear transformation. The nonlinear transformation uses a two-layer neural network with 8-dimensional feature supplement input, 6 nodes in the hidden layer, and 4 nodes in the output layer, corresponding to the feature supplement values. After nonlinear transformation, the feature supplement values ​​are obtained [0.15, 0.22, -0.08, 0.17].

[0117] The supplemented feature representation is obtained by adding the supplemented feature value to the selected feature representation. Element-wise addition is performed on the two vectors to obtain the supplemented feature representation [0.93, 1.25, 0.40, 1.20].

[0118] The policy gradient is calculated based on the supplemented feature representation of the initial policy function. The policy gradient represents the partial derivative of the policy function output with respect to the parameters. The supplemented feature representation is used as input. The policy function output is calculated through forward propagation, and the parameter gradient is calculated through backpropagation. For example, the calculated gradient for a certain parameter may be 0.12.

[0119] The policy gradient is multiplied by the pre-set adaptive learning rate to update the parameters in the initial policy function. The adaptive learning rate can be calculated using the Adam optimizer, with an initial value of 0.01, and is dynamically adjusted based on gradient changes. The gradient is multiplied by the learning rate to obtain the parameter update amount, for example, 0.12 × 0.01 = 0.0012. This parameter update amount is then added to the original parameters to complete the parameter update.

[0120] An adaptive threshold for container switching is calculated based on the updated policy parameters. This threshold determines the conditions under which a bitmap indexed container is switched or reconstructed. Using the updated policy function, the current feature representation is taken as input, and a threshold value, such as 0.75, is output. A container switching operation is triggered when the performance metrics of the bitmap indexed container fall below this threshold.

[0121] In this embodiment, by introducing a time decay coefficient sequence to weight historical data and combining it with the L2 norm of feature gradients to calculate importance scores, the key features that truly influence policy decisions can be accurately identified. This not only considers the immediate importance of features but also reflects their importance in the historical evolution process, improving the accuracy of feature selection. Through the interactive operation of query vectors, key-value vectors, and numerical vectors, deep correlations between features are captured. A Gaussian-based random noise vector is introduced to supplement features, and feature supplement values ​​are generated through nonlinear transformations, enhancing the robustness and generalization ability of feature representation. This not only compensates for the information loss that may be caused by feature selection but also improves the model's ability to cope with uncertainty. By using the policy gradient method combined with an adaptive learning rate to update policy parameters, the dynamic adjustment of the container switching threshold is realized, improving the system's adaptability and decision accuracy.

[0122] In one alternative implementation,

[0123] When container type switching is required, a temporary container is created to perform deduplication on the data to be migrated in the bitmap index container. The deduplicated data is then calculated based on its popularity and relevance. The weighted sum of these two values ​​is used as a priority value, and the deduplicated data is sorted to obtain the sequence of data to be migrated, which includes:

[0124] When container type switching is required, a temporary container is created and the data to be migrated from the bitmap index container is input into the temporary container. The feature fingerprint of each data in the data to be migrated is calculated using the locality sensitive hashing algorithm. Based on the feature fingerprint, the data similarity between any two data in the data to be migrated is calculated. The data similarity is compared with a preset similarity threshold. When the data similarity is greater than the preset similarity threshold, it is determined to be duplicate data and removed to obtain deduplicated data.

[0125] The access frequency is obtained by counting the number of times the deduplicated data is accessed within the second sliding time window. The time weight is obtained by calculating the time interval between the last access and the current time. The survival time is obtained by calculating the time interval between the data generation time and the current time. The popularity value is calculated based on the access frequency, the time weight and the survival time.

[0126] The association weight is obtained by calculating the reference relationship between each pair of data in the deduplicated data, the semantic similarity between each pair of data is calculated, and the association weight and the semantic similarity are multiplied and the average value is calculated to obtain the association degree value.

[0127] According to the task requirements, the weight coefficients corresponding to the popularity value and the relevance value are initialized and weighted summed to obtain the priority value. Based on the priority value, the deduplicated data is sorted in descending order to obtain the data sequence to be migrated.

[0128] When the performance of the bitmap index container drops below a preset threshold, or when an adaptive strategy determines that a container type switch is necessary, a temporary container is created, and the data to be migrated from the bitmap index container is input into the temporary container. The temporary container can be a storage structure with the same structure as the target container type but a smaller size, used to temporarily store the data to be migrated. For example, when the original bitmap index container is a bitmap structure implemented based on an array, and the target container type is a bitmap index container implemented based on a tree structure, a temporary tree-structured container needs to be created. Assuming the original bitmap index container contains 10,000 data entries, depending on storage space and performance requirements, it may be necessary to migrate all or part of this data to the new container type.

[0129] The feature fingerprint of each data entry in the data to be migrated is calculated using the Locality Sensitive Hash (LSH) algorithm. In this embodiment, the MinHash algorithm is used as the implementation of LSH. For each data entry to be migrated, its key fields are extracted as features, such as the data identifier, type, and content digest. Then, the MinHash algorithm is applied to generate a feature fingerprint. The feature fingerprint is usually a fixed-length integer array, such as an integer array of length 128. For a data record {ID: 1001, Type: "transaction", Content: "payment processing", Timestamp: 1629350400}, the feature fingerprint calculated by the MinHash algorithm may be [45, 78, 32, 91, ..., 64].

[0130] The similarity between any two pieces of data to be migrated is calculated based on feature fingerprints. Data similarity can be measured by the Jaccard similarity coefficient of the two feature fingerprints. The Jaccard similarity coefficient is the intersection of the two sets divided by the union. When using the MinHash algorithm, the Jaccard similarity coefficient can be approximated by calculating the proportion of identical values ​​at the same positions in the two feature fingerprints. For example, for the feature fingerprints [45, 78, 32, 91, ..., 64] and [45, 79, 32, 90, ..., 64], if 112 out of 128 positions have the same value, then their similarity is 112 / 128 = 0.875.

[0131] The data similarity is compared with a preset similarity threshold. Data with a similarity greater than the threshold is considered duplicates and removed, resulting in deduplicated data. The preset similarity threshold can be set according to business needs, for example, 0.85. When the similarity between two data entries exceeds 0.85, they are considered duplicates, and only one is retained. Assuming there were originally 10,000 data entries to be migrated, after deduplication, 8,500 duplicate entries remain.

[0132] The access frequency is calculated by counting the number of times each deduplicated data is accessed within the second sliding time window. This second sliding time window can be set to the most recent 24 hours, counting the number of times each deduplicated data entry is accessed within this time window. Access frequency represents the number of times data is accessed per unit of time, and can be calculated by dividing the number of accesses by the time window length. For example, if a data entry is accessed 120 times in 24 hours, its access frequency is 120 / 24 = 5 times / hour.

[0133] The time weight is calculated by taking the time interval between the last access and the current time. The time weight reflects the timeliness of the data and is calculated using a decreasing function; the shorter the time interval, the greater the time weight. For example, if the last access was 2 hours ago, the time weight is 1 / (2+1) = 0.333; if the last access was 0.5 hours ago, the time weight is 1 / (0.5+1) = 0.667.

[0134] The lifetime is calculated by taking the time interval between the data's creation and the current time. The lifetime represents the length of time the data has existed since its creation. For example, if the data was created 720 hours ago, its lifetime is 720 hours.

[0135] The popularity score is calculated based on access frequency, time weight, and lifetime. The popularity score is calculated by multiplying the access frequency by the time weight and then dividing by the logarithm of the lifetime (adding 1 to prevent the logarithm from being 0). This takes into account the data's access frequency, recent access patterns, and lifetime, providing a comprehensive reflection of the data's popularity. For example, for data with an access frequency of 5 times / hour, a time weight of 0.667, and a lifetime of 720 hours, the popularity score can be calculated as 5 × 0.667 / log(720 + 1) = 1.11.

[0136] The association weight is calculated by determining the reference relationships between every two data items in the deduplicated dataset. Reference relationships can be determined by analyzing explicit references (such as foreign keys) or implicit references (such as operations within the same session). The association weight can be set as the ratio of the number of references between two data items to the total number of references. For example, if data A references data B 30% of all references to A, then the association weight from A to B is 0.3.

[0137] Calculate the semantic similarity between each pair of data. Semantic similarity can be calculated by comparing the content, tags, or attributes of the data. Cosine similarity, Word2Vec, or other semantic analysis methods can be used. For example, for the content vectors of the two data points [0.2, 0.5, 0.3, 0.8] and [0.3, 0.4, 0.2, 0.7], the calculated cosine similarity is 0.92.

[0138] The association score is obtained by multiplying the association weight by the semantic similarity and then averaging the results. The association score comprehensively considers both the referencing relationship and semantic similarity between the data. For example, for two data points with an association weight of 0.3 and a semantic similarity of 0.92, their association score is 0.3 × 0.92 = 0.276. The association score between a data point and all other data points is averaged to obtain the overall association score for that data point.

[0139] The priority value is obtained by initializing the weight coefficients corresponding to the popularity and relevance values ​​according to the task requirements and then summing them by weight. Different tasks place different emphasis on popularity and relevance, which is reflected by setting different weight coefficients. For example, for tasks that emphasize data popularity, the popularity weight can be set to 0.7 and the relevance weight to 0.3; for tasks that emphasize data relevance, the popularity weight can be set to 0.3 and the relevance weight to 0.7. The priority value is calculated by multiplying the popularity value by its popularity weight and then adding the relevance value by its relevance weight. For example, for data with a popularity value of 1.11 and a relevance value of 0.276, if the popularity weight is 0.7 and the relevance weight is 0.3, the priority value is 1.11 × 0.7 + 0.276 × 0.3 = 0.86.

[0140] The deduplicated data is sorted in descending order based on priority values ​​to obtain the sequence of data to be migrated. Higher priority values ​​indicate more important data that should be migrated first. All deduplicated data is then sorted from highest to lowest priority value to form the sequence of data to be migrated. For example, the priority values ​​of the first three data entries after sorting might be 2.56, 2.34, and 2.15, respectively. In the actual migration process, data can be migrated sequentially according to this sequence to ensure that important data is processed first, improving overall system performance.

[0141] In this embodiment, a temporary container is created when switching container types. Locality-Sensitive Hash (LSH) algorithm is used to generate feature fingerprints for the data to be migrated, calculate data similarity, and remove duplicate data, thereby reducing redundant data and migration overhead. The association weight is obtained by calculating the reference relationship between data, and the association degree value is obtained by combining semantic similarity. This achieves an accurate assessment of the structural relationship and semantic relevance between data. According to the task requirements, the weight coefficients of the heat value and the association degree value are initialized and weighted summed to obtain the priority value. The deduplicated data is sorted in descending order of priority to generate a sequence of data to be migrated. Thus, during the migration process, the effect of prioritizing the transmission of high-value data while ensuring the integrity of the association between data is achieved. This effectively reduces the amount of data migration, shortens the migration time, and improves the overall performance and business continuity of the container switching process.

[0142] In one alternative implementation,

[0143] The data sequence to be migrated is written to a temporary container in batches, and the write status value is recorded. When the write status value indicates that the pre-commit was successful, a confirmation commit is executed, a bidirectional index table is generated, and the bidirectional index table is used to redirect data access requests, including:

[0144] The data sequence to be migrated is grouped into multiple data batches according to a preset batch size. The multiple data batches are written into a temporary container, and a status value is recorded for each data batch.

[0145] Calculate the hash value of the data in each data batch and sum them to obtain the data checksum. Compare the data checksum after writing with the data checksum before writing to obtain the integrity verification result. Generate a pre-commit log record based on the integrity verification result. When the write status value indicates that the pre-commit is successful, execute the confirmation commit operation.

[0146] Obtain the original data identifier of the data to be migrated from the bitmap index container, obtain the new data identifier of the data to be migrated from the temporary container, establish a mapping relationship and add version number and timestamp to generate metadata information, and construct a bidirectional index table based on the original data identifier, the new data identifier and the metadata information;

[0147] The mapping relationship between the original data identifier and the new data identifier is stored in the redirection cache. When a data access request is received, the target data location is first searched in the redirection cache. If the target data location is not found, the target data location is searched in the bidirectional index table. If the search fails, it is determined whether the number of retries exceeds the search threshold. If it does not exceed the threshold, a retry is performed. If it exceeds the threshold and the operation is a read operation, the data in the bitmap index container is accessed. If the threshold is exceeded and the operation is a write operation, the data access request is rejected.

[0148] For a sorted sequence of data to be migrated, it is grouped into multiple batches according to a preset batch size. The preset batch size is a parameter set based on system performance and fault tolerance requirements, typically ranging from 100 to 1000 data entries. For example, for 8500 deduplicated data entries, setting the batch size to 500 entries will result in 17 batches. The original priority order is maintained during grouping; the first batch contains the 500 highest-priority data entries, the second batch contains the next highest-priority 500 data entries, and so on.

[0149] Multiple data batches are written to a temporary container, and a write status value is recorded for each data batch. The write operation converts each piece of data in the batch into a format supported by the temporary container and stores it according to the data organization method of the temporary container. The write status value is used to identify the write status of the data batch, including "initialization", "writing in progress", "write complete", "pre-commit successful", "confirmed commit complete", or "write failed". For example, for the first batch of 500 data, the status is set to "writing in progress" at the start of the write and updated to "write complete" after the write is completed.

[0150] Calculate the hash value of each data entry in each batch and sum them to obtain the data checksum. Hash value calculation can use common hash algorithms such as SHA-256 or MD5. For each data entry in the batch, calculate its hash value, then add all hash values ​​together (or perform an XOR operation) to obtain the data checksum. For example, for a batch containing 500 data entries, the data checksum calculated before writing might be 0x8F72A1B3C5D4E6F7.

[0151] The integrity verification result is obtained by comparing the checksum of the data after writing with the checksum of the data before writing. After the data is written to the temporary container, the data checksum is calculated again and compared with the checksum before writing. If the two checksums are equal, the data integrity verification is successful; otherwise, it indicates that data corruption occurred during the writing process. For example, if the checksum of the data after writing is also 0x8F72A1B3C5D4E6F7, the integrity verification result is "passed".

[0152] A pre-commit log record is generated based on the integrity verification result. The pre-commit log record contains information such as batch ID, write time, number of data entries, checksum, and verification result. For example, for the first batch, the pre-commit log record might be {BatchID: 1, Time: "2023-08-10 14:30:25", Count: 500, Checksum: "0x8F72A1B3C5D4E6F7", Validation: "Passed"}. A confirmation commit operation is performed when the write status value indicates a successful pre-commit. The confirmation commit operation updates the current batch's status from "Pre-commit successful" to "Confirm commit complete," indicating that the batch data has been safely written to the temporary container.

[0153] Retrieve the original data identifier of the data to be migrated from the bitmap index container. The original data identifier is a unique identifier for the data in the original bitmap index container, which may be an integer ID or a hash value. For example, the identifier of a piece of data to be migrated in the original bitmap index container is "BM_123456".

[0154] Retrieve a new data identifier from the temporary container for the data to be migrated. This new data identifier is a unique identifier for the data within the temporary container (the future new container). For example, the identifier for the same piece of data in the temporary container might be "TR_789012".

[0155] Establish a mapping relationship and add a version number and timestamp to generate metadata information. The mapping relationship records the correspondence between the original data identifier and the new data identifier. The version number is used to identify the version of the data, initially set to 1. The timestamp records the time when the mapping relationship was created. Metadata information may be like {OriginalID: "BM_123456", NewID: "TR_789012", Version: 1, Timestamp: 1628590225000}.

[0156] A bidirectional index table is constructed based on the original data identifier, the new data identifier, and metadata information. The bidirectional index table contains mappings in two directions: from the original data identifier to the new data identifier, and from the new data identifier back to the original data identifier. The bidirectional index structure supports fast lookups and reverse lookups. For example, a record in the bidirectional index table might be {Forward: {Key: "BM_123456", Value: "TR_789012"}, Backward: {Key: "TR_789012", Value: "BM_123456"}, Metadata: {Version: 1, Timestamp: 1628590225000}}.

[0157] The mapping relationship between the original data identifier and the new data identifier is stored in the redirection cache. The redirection cache is a high-speed cache structure used to accelerate the data location lookup process. The cache is managed using the LRU (Least Recently Used) strategy, and its capacity can be set to 10,000 most recently accessed mapping relationships. For example, an entry in the redirection cache might be {Key: "BM_123456", Value: "TR_789012", AccessTime: 1628590230000}.

[0158] Upon receiving a data access request, the system first searches for the target data location in the redirection cache. The data access request contains the identifier of the target data and the operation type (read or write). The search process checks the redirection cache to determine if the requested data identifier exists in the cache. For example, upon receiving a request to read the data "BM_123456", the system searches the redirection cache for an entry with the key "BM_123456".

[0159] If not found, the target data location is searched in the bidirectional index table. If there is no relevant mapping in the redirection cache, the search continues in the bidirectional index table. For example, searching for the entry with the key "BM_123456" in the bidirectional index table will find the corresponding new data identifier "TR_789012".

[0160] If the search fails, check if the number of retries exceeds the search threshold. Search failures may be due to data not yet being migrated or the mapping relationship not yet being established. Configure a retry mechanism to allow multiple search attempts, with a retry interval of 10 milliseconds and a search threshold of 3 attempts. For example, after the first search fails, wait 10 milliseconds before searching again, until the data is found or the retry limit is reached.

[0161] If the number of retries does not exceed the threshold, a retry is performed. If the threshold is exceeded and the operation type is read, the data in the bitmap index container is accessed. If the number of retries does not exceed the threshold, the search continues in the redirect cache and bidirectional index table. If the threshold is exceeded and the operation type is read, the data is read back from the original bitmap index container. For example, if all three lookups fail and the request is a read operation, the data corresponding to "BM_123456" is read directly from the original bitmap index container.

[0162] If the number of retries exceeds the threshold and the operation type is write, the data access request will be rejected. If the number of retries exceeds the threshold and the operation type is write, the write request will be rejected and the error message "Data migration in progress, write operation is not supported at this time" will be returned.

[0163] In this embodiment, hash value summation is introduced during the migration process to generate a data checksum, which is then compared with the checksum before writing. This achieves integrity verification of the data batch, ensuring the accuracy and reliability of the migrated data during the writing process. The pre-commit log mechanism confirms the submission only after verification, effectively reducing the risk of data loss or corruption due to writing errors. By obtaining the original data identifier and the new data identifier and establishing a mapping relationship, combined with the version number and timestamp to generate metadata information and build a bidirectional index table, the data location can be quickly located after migration and forward and reverse traceability can be supported, ensuring data consistency and manageability. The bidirectional index table is used as a secondary retrieval mechanism to improve the success rate of the search, and the retry mechanism enhances the robustness of access. After exceeding the search count threshold, read rollback or write rejection is performed according to the operation type, thereby achieving efficient access control and fault tolerance during the migration process while ensuring data consistency and integrity.

[0164] Figure 2 This is a data migration logic flowchart of the bitmap index container optimization method based on multi-dimensional dynamic decision-making in an embodiment of the present invention.

[0165] In one alternative implementation,

[0166] Monitor performance metrics after the switchover. When a performance metric exceeds a preset range, perform a rollback operation and update decision parameters, including:

[0167] Obtain the performance metrics after the switch, including service response time, resource utilization, and service success rate;

[0168] Determine whether the performance index value exceeds a preset performance threshold range. The preset performance threshold range is determined based on the statistical distribution of historical performance data. When the duration for which the performance index value continuously exceeds the preset performance threshold range is greater than a preset time threshold, a rollback operation is triggered.

[0169] The rollback operation is executed and the decision parameters are updated based on the performance fluctuations of this switch. The decision parameters include the switch time window, performance monitoring frequency, and performance threshold range.

[0170] After the bitmap index container switch is completed, the system performance is continuously monitored to promptly identify anomalies and take corresponding measures. Performance metrics obtained after the switch include service response time, resource utilization, and business success rate. Service response time measures the time required from receiving a request to returning a result, measured in milliseconds. Resource utilization reflects the usage of system resources (such as CPU, memory, disk I / O, etc.), expressed as a percentage. Business success rate represents the proportion of successfully processed business requests out of the total number of requests, also expressed as a percentage. Performance metrics can be collected in real time through a performance monitoring module embedded in the bitmap index container or obtained through integration with an external monitoring system. For example, in the first hour after the switch, performance data is collected every minute, resulting in the following sequence of metrics: service response time [15ms, 16ms, 14ms, 17ms, ...], resource utilization [45%, 47%, 46%, 48%, ...], and business success rate [99.8%, 99.7%, 99.9%, 99.8%, ...].

[0171] The system determines whether performance metric values ​​exceed preset performance thresholds. These thresholds are determined based on the statistical distribution of historical performance data, typically using the mean plus or minus a certain number of standard deviations to set the upper and lower limits. For example, for service response time, assuming a historical mean of 12ms and a standard deviation of 2ms, the threshold range could be set to [6ms, 18ms] (mean plus or minus 3 times the standard deviation). For resource utilization, assuming a historical mean of 40% and a standard deviation of 5%, the threshold range could be set to [25%, 55%]. For business success rate, since a high target is desired, only a lower limit can be set, such as 99.5%. When collected performance metric values ​​are compared with the preset threshold ranges, it determines whether there are any instances of consecutively exceeding the thresholds. Suppose that during monitoring, the service response time is found to be [19ms, 20ms, 21ms, 22ms, 23ms] consecutively, all exceeding the preset upper limit of 18ms; or the business success rate is found to be [99.4%, 99.3%, 99.2%, 99.3%, 99.4%] consecutively, all below the preset lower limit of 99.5%.

[0172] A rollback operation will be triggered when a performance metric value exceeds a preset performance threshold for a duration greater than a preset time threshold. The preset time threshold is a time period set based on business tolerance and system stability requirements, typically 5-15 minutes. For example, if the preset time threshold is set to 5 minutes and the monitoring frequency is once per minute, then a rollback operation will be triggered when the performance metric value exceeds the threshold for 5 consecutive monitoring sessions.

[0173] The rollback operation involves stopping the writing of new data to the new container, retrieving the mapping relationship of the migrated data from the bidirectional index table, writing the data back to the original bitmap index container according to the mapping relationship, updating routing information, and redirecting access requests back to the original bitmap index container. To ensure the reliability of the rollback process, batch processing is used, along with verification and comparison. For example, for 8500 migrated data entries, they are written back in batches of 500, with verification and comparison performed after each batch to ensure data consistency.

[0174] The decision parameters are updated based on the performance fluctuations observed during this switchover. These parameters include the switchover time window, performance monitoring frequency, and performance threshold range. The switchover time window refers to the optimal time period for performing the container switchover, typically chosen during periods of low system load. For example, if the switchover is performed during a period of high system load and causes performance issues, the switchover time window needs to be adjusted to a period of lower load, such as changing it from a weekday afternoon to nighttime or weekends. The performance monitoring frequency refers to the time interval for collecting performance metrics, which needs to be adjusted based on the system's response speed and fluctuation characteristics. For example, if performance fluctuations are significant after the switchover, the monitoring frequency needs to be increased from once per minute to once every 30 seconds to detect problems more promptly. The performance threshold range is the standard for judging whether system performance is abnormal and needs to be dynamically adjusted based on the actual situation. For example, if the service response time fluctuation range is larger than expected after the switchover, the response time threshold range may need to be adjusted from [6ms, 18ms] to [5ms, 20ms] to reduce false positives.

[0175] Decision parameters are updated based on the statistical characteristics of performance fluctuations. For the switching time window, the system load in different time periods is analyzed, and the period with the lowest load is selected as the new switching time window. For example, by analyzing the system load every hour over a week, it is found that the load is lowest between 2 AM and 4 AM, so this period is set as the new switching time window. The performance monitoring frequency can be adjusted according to the rate of change of performance indicators. For example, if the rate of change of performance indicators is high, it indicates that the system state is changing rapidly, and the monitoring frequency needs to be increased; conversely, if the rate of change is low, the monitoring frequency can be appropriately reduced to decrease system overhead. The performance threshold range can be updated based on the statistical distribution of historical data. For example, if the mean of the performance indicator over a recent period is 14ms and the standard deviation is 3ms, the response time threshold range can be updated to [5ms, 23ms] (mean plus or minus 3 times the standard deviation).

[0176] In this embodiment, performance metrics such as service response time, resource utilization, and business success rate are acquired in real time after container type switching, and compared with a preset performance threshold range determined based on the statistical distribution of historical performance data. When a performance metric value is detected to continuously exceed the threshold range for a duration exceeding a preset time threshold, a rollback operation is automatically triggered. This allows for rapid recovery to a stable state during performance anomalies, reducing the impact of switching on business continuity and user experience. After rollback, the switching decision parameters are dynamically updated based on the performance fluctuations of this switch, enabling adaptive optimization of subsequent switching strategies, continuously improving the accuracy and stability of switching decisions, and achieving proactive prevention and continuous optimization of performance risks.

[0177] A second aspect of the present invention provides a bitmap index container optimization system based on multidimensional dynamic decision-making, comprising:

[0178] The first unit is used to obtain the data feature values ​​of the bitmap index container and calculate the skewness coefficient value based on the data feature values;

[0179] The second unit is used to process data feature values ​​through a pre-set time series diagram structure to obtain a data distribution feature vector, determine a data access pattern vector and a corresponding data change trend vector based on the data feature values, combine the data distribution feature vector, data access pattern vector and data change trend vector across dimensions to obtain a feature fusion vector, and iteratively optimize the decision parameters based on the feature fusion vector and a dual-strategy gradient algorithm to obtain an adaptive threshold for container switching.

[0180] The third unit is used to compare the skewness coefficient value with the adaptive threshold and determine whether to switch the container type.

[0181] The fourth unit is used to create a temporary container to perform deduplication on the data to be migrated in the bitmap index container when container type switching is required. It calculates the popularity and relevance values ​​of the deduplicated data, uses the weighted sum of the popularity and relevance values ​​as the priority value, sorts the deduplicated data to obtain the sequence of data to be migrated, writes the sequence of data to be migrated into the temporary container in batches and records the write status value. When the write status value indicates that the pre-commit is successful, it performs a confirmation commit, generates a bidirectional index table, and uses the bidirectional index table to redirect data access requests.

[0182] The fifth unit is used to monitor the performance index values ​​after the switch. When the performance index values ​​exceed the preset range, a rollback operation is performed and the decision parameters are updated.

[0183] A third aspect of the present invention provides an electronic device, comprising:

[0184] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0185] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0186] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0187] 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; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to 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 bitmap index container optimization method based on multidimensional dynamic decision-making, characterized in that, include: Obtain the data feature values ​​of the bitmap index container and calculate the skewness coefficient value based on the data feature values; Data distribution feature vectors are obtained by processing data feature values ​​through a pre-set time series diagram structure. Data access pattern vectors and corresponding data change trend vectors are determined based on the data feature values. Data distribution feature vectors, data access pattern vectors, and data change trend vectors are combined across dimensions to obtain feature fusion vectors. Based on the feature fusion vectors, decision parameters are iteratively optimized using a dual-strategy gradient algorithm to obtain the adaptive threshold for container switching. The skewness coefficient value is compared with the adaptive threshold to determine whether to switch the container type. When container type switching is required, a temporary container is created to perform deduplication on the data to be migrated in the bitmap index container to obtain deduplicated data. The popularity value and relevance value of the deduplicated data are calculated. The weighted sum of the popularity value and relevance value is used as the priority value and the deduplicated data is sorted to obtain the sequence of data to be migrated. The sequence of data to be migrated is written to the temporary container in batches and the write status value is recorded. When the write status value indicates that the pre-commit is successful, the confirmation commit is executed, a bidirectional index table is generated, and the bidirectional index table is used to redirect data access requests. Monitor the performance metrics after the switch, and when the performance metrics exceed the preset range, perform a rollback operation and update the decision parameters. Data distribution feature vectors are obtained by processing data feature values ​​through a pre-set time series graph structure. Based on the data feature values, data access pattern vectors and corresponding data change trend vectors are determined. The feature fusion vector is obtained by combining the data distribution feature vector, data access pattern vector, and data change trend vector across dimensions. A time series graph structure is constructed based on the data feature values ​​based on a pre-set time interval and multiple time nodes are recorded. The bitmap cell state parameters of each time node are calculated. A state transition matrix is ​​constructed based on the bitmap cell state parameters and feature decomposition is performed to obtain the data distribution feature vector. The bitmap cell state parameters include occupancy status, modification time and index information. Data blocks in bitmap units are obtained from the time sequence diagram structure. The standard access count of the data blocks within a preset time period and the consecutive access count between the data blocks are recorded. The ratio of the consecutive access count to the standard access count is used to construct an access pattern matrix. The standardized Laplace matrix of the access pattern matrix is ​​calculated and the feature vector is extracted as a data access pattern vector. The time window sequence is divided according to a preset time length. The number of data operations in each time window is recorded. The ratio of the number of data operations to the total number of operations in the time window is calculated to obtain the operation ratio parameter. The operation ratio parameter is multiplied by the corresponding preset weight coefficient and summed to obtain the trend index. The trend index is subjected to time series analysis to obtain the data trend vector. The number of data operations includes the number of data writes, the number of data updates, and the number of data deletions. The data distribution feature vector, data access pattern vector, and data change trend vector are dimensionally aligned and normalized, and the results are weighted and summed to obtain the feature fusion vector.

2. The method according to claim 1, characterized in that, Obtaining the data feature values ​​of a bitmap indexed container and calculating the skewness coefficient value based on the data feature values ​​includes: Within a preset first time window, the number of writes and reads of bitmap cells in the bitmap index container is counted, the distribution density of data storage locations in the bitmap cells is calculated, and data feature values ​​are obtained. The data feature values ​​are divided into multiple data segments according to preset rules, and the distribution characteristics of each data segment are calculated. The segments are then combined to obtain a probability distribution sequence. The skewness coefficient value corresponding to the bitmap index container is calculated based on the probability distribution sequence.

3. The method according to claim 1, characterized in that, Based on the aforementioned feature fusion vector, and by iteratively optimizing the decision parameters using a dual-policy gradient algorithm, the adaptive threshold for container switching is obtained, including: Historical data is acquired to construct a historical data sequence. Based on the length of the historical data sequence, a time decay coefficient sequence is constructed as a time decay weight. The gradient of each feature in the feature fusion vector with respect to the pre-set initial policy function is calculated. The L2 norm of the gradient is multiplied by the time decay weight and accumulated to obtain the importance score of each feature. Based on the importance score, an attention transformation matrix group is constructed. The feature fusion vector is multiplied by the attention transformation matrix group to obtain the query vector, key vector, and numerical vector. The dot product of the query vector and the key vector is calculated, and after softmax normalization, it is multiplied by the numerical vector to obtain the reconstructed feature representation. One-dimensional convolution operation and ReLU function activation are performed on the reconstructed feature representation. The compressed feature representation is obtained by mapping through a fully connected layer. The feature selection mask is determined by the sigmoid function and multiplied element-wise with the compressed feature representation to obtain the selected feature representation. A random noise vector is constructed based on a Gaussian distribution. The selected feature representation is connected to the random noise vector to form a feature supplement input. A feature supplement value is generated by combining nonlinear transformation. The feature supplement value is added to the selected feature representation to obtain the supplemented feature representation. The policy gradient corresponding to the supplemented feature representation is calculated based on the initial policy function. The policy gradient is multiplied by the pre-set adaptive learning rate to update the parameters in the initial policy function, and the adaptive threshold for container switching is calculated based on the updated policy parameters.

4. The method according to claim 1, characterized in that, When container type switching is required, a temporary container is created to perform deduplication on the data to be migrated in the bitmap index container. The deduplicated data is then calculated based on its popularity and relevance. The weighted sum of these two values ​​is used as a priority value, and the deduplicated data is sorted to obtain the sequence of data to be migrated, which includes: When container type switching is required, a temporary container is created and the data to be migrated from the bitmap index container is input into the temporary container. The feature fingerprint of each data in the data to be migrated is calculated using the locality sensitive hashing algorithm. Based on the feature fingerprint, the data similarity between any two data in the data to be migrated is calculated. The data similarity is compared with a preset similarity threshold. When the data similarity is greater than the preset similarity threshold, it is determined to be duplicate data and removed to obtain deduplicated data. The access frequency is obtained by counting the number of times the deduplicated data is accessed within the second sliding time window. The time weight is obtained by calculating the time interval between the last access and the current time. The survival time is obtained by calculating the time interval between the data generation time and the current time. The popularity value is calculated based on the access frequency, the time weight and the survival time. The association weight is obtained by calculating the reference relationship between each pair of data in the deduplicated data, the semantic similarity between each pair of data is calculated, and the association weight and the semantic similarity are multiplied and the average value is calculated to obtain the association degree value. According to the task requirements, the weight coefficients corresponding to the popularity value and the relevance value are initialized and weighted summed to obtain the priority value. Based on the priority value, the deduplicated data is sorted in descending order to obtain the data sequence to be migrated.

5. The method according to claim 1, characterized in that, The data sequence to be migrated is written to a temporary container in batches, and the write status value is recorded. When the write status value indicates that the pre-commit was successful, a confirmation commit is executed, a bidirectional index table is generated, and the bidirectional index table is used to redirect data access requests, including: The data sequence to be migrated is grouped into multiple data batches according to a preset batch size. The multiple data batches are written into a temporary container, and a status value is recorded for each data batch. Calculate the hash value of the data in each data batch and sum them to obtain the data checksum. Compare the data checksum after writing with the data checksum before writing to obtain the integrity verification result. Generate a pre-commit log record based on the integrity verification result. When the write status value indicates that the pre-commit is successful, execute the confirmation commit operation. Obtain the original data identifier of the data to be migrated from the bitmap index container, obtain the new data identifier of the data to be migrated from the temporary container, establish a mapping relationship and add version number and timestamp to generate metadata information, and construct a bidirectional index table based on the original data identifier, the new data identifier and the metadata information; The mapping relationship between the original data identifier and the new data identifier is stored in the redirection cache. When a data access request is received, the target data location is first searched in the redirection cache. If the target data location is not found, the target data location is searched in the bidirectional index table. If the search fails, it is determined whether the number of retries exceeds the search threshold. If it does not exceed the threshold, a retry is performed. If it exceeds the threshold and the operation is a read operation, the data in the bitmap index container is accessed. If the threshold is exceeded and the operation is a write operation, the data access request is rejected.

6. The method according to claim 1, characterized in that, Monitor performance metrics after the switchover. When a performance metric exceeds a preset range, perform a rollback operation and update decision parameters, including: Obtain the performance metrics after the switch, including service response time, resource utilization, and service success rate; Determine whether the performance index value exceeds a preset performance threshold range. The preset performance threshold range is determined based on the statistical distribution of historical performance data. When the duration for which the performance index value continuously exceeds the preset performance threshold range is greater than a preset time threshold, a rollback operation is triggered. The rollback operation is executed and the decision parameters are updated based on the performance fluctuations of this switch. The decision parameters include the switch time window, performance monitoring frequency, and performance threshold range.

7. A bitmap index container optimization system based on multidimensional dynamic decision-making, used to implement the method of any one of claims 1-6, characterized in that, include: The first unit is used to obtain the data feature values ​​of the bitmap index container and calculate the skewness coefficient value based on the data feature values; The second unit is used to process data feature values ​​through a pre-set time series diagram structure to obtain a data distribution feature vector, determine a data access pattern vector and a corresponding data change trend vector based on the data feature values, combine the data distribution feature vector, data access pattern vector and data change trend vector across dimensions to obtain a feature fusion vector, and iteratively optimize the decision parameters based on the feature fusion vector and a dual-strategy gradient algorithm to obtain an adaptive threshold for container switching. The third unit is used to compare the skewness coefficient value with the adaptive threshold and determine whether to switch the container type. The fourth unit is used to create a temporary container to perform deduplication on the data to be migrated in the bitmap index container when container type switching is required. It calculates the popularity and relevance values ​​of the deduplicated data, uses the weighted sum of the popularity and relevance values ​​as the priority value, sorts the deduplicated data to obtain the sequence of data to be migrated, writes the sequence of data to be migrated into the temporary container in batches and records the write status value. When the write status value indicates that the pre-commit is successful, it performs a confirmation commit, generates a bidirectional index table, and uses the bidirectional index table to redirect data access requests. The fifth unit is used to monitor the performance index values ​​after the switch. When the performance index values ​​exceed the preset range, a rollback operation is performed and the decision parameters are updated.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images