Log structure storage data placement method
By using dynamic grouping and cross-group aggregation mechanisms, combined with access density-aware threshold adaptation and proactive degradation placement strategies, the write amplification problem of log-structured storage systems under sparse access loads is solved, thereby improving the performance of the storage system and the lifespan of the hard drives.
Patent Information
- Application Number
- CN202511346147.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-19
- Publication Date
- 2026-01-16
AI Technical Summary
Existing log-structured storage systems suffer from excessive write amplification under sparse access loads, leading to performance degradation and shortened SSD lifespan. Existing data placement algorithms have failed to effectively address the high write costs and write amplification caused by granularity mismatch.
By using dynamic grouping and cross-group aggregation mechanisms, data blocks are classified into hot data blocks and cold data blocks. Hot data blocks are stored in the unfilled space of cold groups. The storage mechanism is optimized through delayed storage, combined with access density-aware threshold adaptation and proactive degradation placement strategies to reduce unnecessary migration and filling operations.
It effectively reduces write amplification, improves the throughput and write efficiency of the storage system, extends the lifespan of solid-state drives, adapts to different workload modes, and optimizes the storage and recycling process of data blocks.
Smart Images
Figure CN121349362A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of data storage technology, and more specifically, relates to a method for placing data in a log structure storage. Background Technology
[0002] With the continuous development of information technology, solid-state drives (SSDs) have become the primary choice for modern storage systems, especially in scenarios with increasingly demanding high-performance data storage and access requirements. SSDs are widely used in various data storage platforms due to their high read / write speeds and low latency. However, the storage characteristics of SSDs differ significantly from traditional disk storage, particularly during garbage collection, where write amplification becomes particularly prominent. This leads to a substantial decrease in SSD performance and lifespan over prolonged use. Log-structured storage architecture is a common design for SSDs, effectively improving performance in high-concurrency write scenarios by appending data to log files. However, garbage collection is an unavoidable process in log-structured storage architectures. Its purpose is to reclaim invalid data blocks and reorganize storage space, but this process often results in a large number of data migrations and write operations, leading to write amplification, increasing unnecessary write burdens, and reducing SSD efficiency and lifespan.
[0003] To mitigate write amplification in log-structured storage, current solutions focus on designing data placement strategies by grouping data into segments and placing data blocks in independent groups. For example, they attempt to model access frequency and recent preference based on access load, or use methods such as clustering for feature-based data classification and placement. To this end, these designs typically divide accesses into several streams, and the more streams, the better the effect. However, these designs incur high write costs on SSD arrays. This is due to the mismatch between the log-structured storage block size and the storage array block size. Log-structured storage is typically organized and managed with a granularity of 4KB pages, while storage arrays often have a larger block granularity (64KB). Thus, when the log-structured storage system frequently sends small-granularity requests directly to the storage array, it results in additional read-modify-write overhead because the storage array blocks cannot be filled. This overhead increases the actual write amplification rate and significantly impacts the overall performance and hardware lifespan of the storage array. Therefore, there is an urgent need for a method that can dynamically adjust data placement strategies, improve garbage collection efficiency, and reduce write amplification. Summary of the Invention In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a log structure storage data placement method, which aims to solve the problem of excessive write amplification in the existing data placement algorithm under sparse access load, so as to avoid the degradation of performance and solid-state drive life.
[0004] To achieve the above objectives, according to one aspect of the present invention, a method for placing log structure storage data is provided, comprising: S1. Based on the access frequency and popularity of the data block corresponding to each write request in the user write request stream, classify the data block as a hot data block or a cold data block, and group it into a hot data group or a cold data group accordingly. S2. Calculate the proportion of zero-padding data blocks for each data group during the persistence process in real time. When the proportion of zero-padding data blocks for both the hot and cold data groups exceeds a preset threshold, cross-group aggregation is triggered. The cross-group aggregation method is as follows: the hot data block currently being persisted corresponding to the current user load in the hot data group is recorded as cross-group data; the cold data block currently being persisted corresponding to the current user load and the cross-group data constitute aggregated data. If the current aggregated data size is aligned with the large block in the storage array, and / or the aggregated data persistence timer times out, then the aggregated data is persisted, thus realizing the cross-group data... Temporary storage; the hot data group persistence processing timer is reset, and new hot data blocks obtained from the new classification are received. After the hot data group persistence processing timer expires, the most recent cross-group data in the hot data group and the newly received hot data blocks after it are persisted together, realizing the delayed persistence of the most recent cross-group data. After this persistence processing, the most recent cross-group data is deleted from the aggregated data persistence. If the cross-group aggregation condition is not met, each data group is persisted separately after the corresponding persistence timer expires, and finally the placement and management of log structure storage data is realized.
[0005] Furthermore, the access frequency and popularity are characterized by the most recent access interval t of the data block; The implementation method for classifying data blocks is as follows: The most recent access interval t of the data block is compared with the optimal classification threshold of the log structure storage system. If t is greater than the optimal classification threshold, the data block is classified as a cold data block; otherwise, the data block is classified as a hot data block.
[0006] Furthermore, the optimal classification threshold is dynamically adjusted in the following manner: The user write request stream is sampled, and a portion of the write requests are randomly selected from the write request stream at a preset fixed sampling rate to determine the data block corresponding to each sampled write request.
[0007] Simulations are performed on each of the identified data blocks using M preset ghost sets: the most recent access interval for each data block is calculated, and the most recent access interval is compared with the hot / cold separation threshold for each ghost set. Compare the intervals; if the interval between the most recent accesses is greater than... If the write amplification is positive, the data block is assigned to the cold data group; otherwise, the data block is assigned to the hot data group. Each ghost set runs the garbage collection process independently and records its write amplification performance. The write amplification performance of all ghost sets is compared, and the cold / hot separation threshold corresponding to the minimum write amplification is selected as the optimal separation threshold for the log structure storage system.
[0008] Furthermore, the M hot and cold separation thresholds are initialized according to an exponential growth relationship; Once an optimal separation threshold is found, the subsequent sampling process adjusts the M hot and cold separation thresholds according to a linear relationship; when the cumulative number of write requests by users exceeds 10% of the storage capacity or the write amplification tends to stabilize, the hot and cold separation thresholds are triggered to be reinitialized.
[0009] Furthermore, the proportion of zero-padding data blocks The method for determining it is as follows:
[0010] In the formula, Represents data group i The number of data blocks already placed inside, data groups i Take cold data set and hot data set, Indicates the size of each data block. This indicates the number of data blocks contained within a segment in a log-structured storage system. It is a data group i Number of zero-filling times k Represents data group i The number of complete segments formed by the internal data blocks.
[0011] Furthermore, prior to S1, the method also includes: For each data block corresponding to a user write request, query the pre-built re-access discriminator. If a match is found, the data block is directly migrated to the garbage collection write group of the corresponding lifecycle; otherwise, execute S1. The re-access discriminator is constructed in the following way: When migrating valid data blocks during garbage collection, if a data block in the selected collection segment is re-identified according to its lifecycle and then migrated back to the original garbage collection write group, the logical block address of that data block is extracted and saved to the re-access discriminator.
[0012] Furthermore, the re-access discriminator consists of multiple cascaded Bloom filters; The implementation of saving the logical block address to the re-access discriminator is as follows: the current logical block address to be saved is inserted into a Bloom filter; when the data inserted into the Bloom filter reaches the threshold, the Bloom filter is cascaded onto the current cascaded Bloom filter. The query is implemented as follows: a query is performed in all Bloom filters in the re-access discriminator. If the number of Bloom filter riders that can find the data exceeds the threshold, it is considered a hit; otherwise, it is considered a miss.
[0013] According to another aspect of the present invention, an electronic device is provided, including a memory and a processor, the memory storing a computer program, characterized in that the processor executes the computer program to implement the steps of the method described above.
[0014] According to another aspect of the present invention, a computer-readable storage medium is provided, characterized in that the computer-readable storage medium includes a stored computer program, wherein the computer program, when executed by a processor, controls the device where the storage medium is located to perform the steps of the method described above.
[0015] In summary, compared with the prior art, the technical solutions conceived by this invention have the following main advantages: 1. This invention proposes a log structure storage data placement method. First, dynamic grouping is performed, dividing data blocks into hot and cold data blocks based on the access frequency and popularity of the data blocks corresponding to the load. Second, cross-group aggregation is triggered, aggregating user requests based on user load characteristics. During the data block writing process, the unfilled space of the cold group is used to store hot data blocks to avoid frequent filling operations, thereby improving the throughput and write efficiency of the storage system. Finally, optimization is achieved through a delayed storage mechanism, which can effectively solve the problem of excessive write amplification in existing data placement algorithms under sparse access loads, thus avoiding performance degradation and SSD lifespan reduction.
[0016] 2. This invention also proposes an access density-aware threshold adaptation method, which calculates the access interval and lifecycle of each data block to determine an initial hot and cold threshold; after each data access, the hot and cold threshold is dynamically updated based on the access pattern of the data block to adapt to real-time workload and data access density changes; the data block allocation strategy is adjusted according to the updated threshold to ensure load balancing of each group and reduce unnecessary fill operations.
[0017] 3. This invention also proposes a proactive degradation placement strategy. Before classifying data into hot and cold categories, by predicting the future access patterns of data blocks, cold data blocks are migrated to appropriate storage groups in advance, reducing redundant migrations during garbage collection. This method effectively solves the problem of excessive write amplification in existing data placement algorithms under sparse access loads, thereby significantly improving the performance of the storage system and extending the lifespan of solid-state drives. Attached Figure Description
[0018] Figure 1A flowchart illustrating a method for placing log structure storage data according to an embodiment of the present invention; Figure 2 This is an overall framework diagram of a log structure data placement method provided in an embodiment of the present invention; Figure 3 The graph shows the write amplification performance test results of the method of the present invention and other existing solutions on multiple datasets, as provided in the embodiments of the present invention. Figure 4 The graph shows the zero-padding flow test results of the method of the present invention and other existing solutions on multiple datasets, as provided in the embodiments of the present invention. Detailed Implementation
[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.
[0020] This invention provides a data placement method for log structure storage in a solid-state drive array, aiming to solve the problems of high write amplification and large filling overhead caused by granularity mismatch in existing data placement strategies.
[0021] Specifically, this invention reduces write amplification, decreases fill traffic, improves write aggregation efficiency, and significantly enhances system throughput under high concurrency load through three innovative mechanisms: dynamic grouping, cross-group aggregation, and predictive degradation. Among these, predictive degradation is introduced as a preferred approach.
[0022] Example 1 A method for storing data in a log structure, such as Figure 1 As shown, it includes: S1. Based on the access frequency and popularity of the data block corresponding to each write request in the user write request stream, classify the data block as a hot data block or a cold data block, and group it into a hot data group or a cold data group accordingly. S2. Calculate the proportion of zero-padding data blocks for each data group during persistence in real time. When the proportion of zero-padding data blocks for both the hot and cold data groups exceeds a preset threshold, cross-group aggregation is triggered. The cross-group aggregation method is as follows: the hot data block currently being persisted corresponding to the current user load in the hot data group is recorded as cross-group data; the cold data block currently being persisted corresponding to the current user load and the cross-group data constitute aggregated data. If the current aggregated data size is aligned with the large block in the storage array (i.e., the solid-state drive storage array), and / or the aggregated data persistence timer times out, then the aggregated data is persisted. The system temporarily stores cross-group data; it resets the persistence timer for hot data groups, continues to receive hot data blocks obtained from new classifications, and persists the most recent cross-group data in the hot data group along with newly received hot data blocks after it expires, thus achieving delayed persistence of the most recent cross-group data. After this persistence process, the most recent cross-group data is deleted from the aggregated data persistence. If the cross-group aggregation conditions are not met, each data group is persisted separately after the corresponding persistence timer expires, ultimately achieving the placement and management of log structure storage data.
[0023] In this embodiment, S1 corresponds to dynamic grouping, which dynamically divides data blocks into hot and cold data blocks based on the access frequency and popularity of the data blocks corresponding to the load. S2 corresponds to cross-group aggregation, which aggregates and places user requests based on user load characteristics. The cross-group aggregation mechanism includes the following steps: checking whether the storage space of the hot data group is sufficient; if not, using the unfilled space of the cold data group to temporarily store the hot data blocks; and performing delayed processing on cross-group data storage, i.e., "lazy appending," to temporarily store the data blocks in the cold group. Therefore, during the data block writing process, S2 utilizes the unfilled space of the cold group to store the hot data blocks, thereby avoiding frequent fill operations and improving the throughput and write efficiency of the storage system. Further optimization is then achieved through the delayed storage mechanism.
[0024] As a preferred implementation, the access frequency and popularity can be characterized by the most recent access interval t of the data block; The implementation method for classifying data blocks is as follows: The most recent access interval t of the data block is compared with the optimal classification threshold of the log structure storage system. If t is greater than the optimal classification threshold, the data block is classified as a cold data block; otherwise, the data block is classified as a hot data block.
[0025] As a preferred approach, monitor user data access patterns and dynamically adjust hot and cold data thresholds to optimize storage space utilization and reduce data migration. Specifically, the optimal classification threshold is dynamically adjusted in the following way: The user write request stream is sampled (for example, the spatial sampling technique uses the SHARDS algorithm), and a portion of the write requests are randomly selected from the write request stream at a preset fixed sampling rate to determine the data block corresponding to each sampled write request.
[0026] Simulations are performed on each of the identified data blocks using M preset ghost sets: the most recent access interval for each data block is calculated, and the most recent access interval is compared with the hot / cold separation threshold for each ghost set. Compare the intervals; if the interval between the most recent accesses is greater than... If the data block is not found to be in the cold data group, it is assigned to the hot data group; otherwise, it is assigned to the hot data group. Each ghost set independently runs the garbage collection process (including victim segment selection, data validity verification, valid data migration, etc.) and records its write amplification performance. The write amplification performance of all ghost sets is compared, and the cold and hot separation threshold corresponding to the minimum write amplification is selected as the optimal separation threshold for the log structure storage system.
[0027] This method is an access density-aware threshold adaptation method, which includes the following steps: calculating the access interval and lifecycle of each data block to determine an initial hot and cold threshold; dynamically updating the hot and cold threshold based on the access pattern of the data block after each data access to adapt to real-time workload and data access density changes; adjusting the data block allocation strategy according to the updated threshold and ensuring load balancing for each group to reduce unnecessary fill operations.
[0028] By dynamically grouping and adjusting hot and cold thresholds, it can effectively adapt to different workload patterns, optimize the storage and reclamation process of data blocks, reduce unnecessary fill traffic, and reduce storage space waste and data migration overhead.
[0029] As a further preferred implementation, the M cold and hot separation thresholds are initialized according to an exponential growth relationship; Once an optimal separation threshold is found, the subsequent sampling process adjusts the M hot and cold separation thresholds according to a linear relationship; when the cumulative number of write requests by users exceeds 10% of the storage capacity or the write amplification tends to stabilize, the hot and cold separation thresholds are triggered to be reinitialized.
[0030] The initial threshold was determined using an exponentially growing sliding window, and subsequently adjusted to a linear growth mode to improve the convergence speed.
[0031] As a further preferred embodiment, the above-mentioned size The method for determining it is as follows:
[0032] In the formula, Represents data group i The number of data blocks already placed inside, data groupsi Take cold data set and hot data set, Indicates the size of each data block. This indicates the number of data blocks contained within a segment in a log-structured storage system. It is a data group i Number of zero-filling times k Represents data group i The number of complete segments formed by the internal data blocks.
[0033] In a preferred embodiment, before S1, the method further includes: A re-access discriminator is used to identify data blocks with long lifecycles: For each data block corresponding to a user write request, the pre-built re-access discriminator is queried. If a match is found, the data block is directly migrated to the garbage collection write group with the corresponding lifecycle; otherwise, S1 is executed. The re-access discriminator is constructed in the following way: When migrating valid data blocks during garbage collection, if a data block in the selected collection segment is re-identified according to its lifecycle and then migrated back to the original garbage collection write group, the logical block address of that data block is extracted and saved to the re-access discriminator.
[0034] In other words, this step corresponds to a proactive degradation placement strategy. During garbage collection, the storage system proactively degrades valid data blocks: by predicting future access patterns of data blocks, cold data blocks are migrated to appropriate storage groups in advance, reducing redundant migrations during garbage collection. This method effectively solves the problem of excessive write amplification in existing data placement algorithms under sparse access loads, thereby significantly improving the performance of the storage system and extending the lifespan of solid-state drives.
[0035] In this implementation, the groups are divided into user write groups and garbage collection write groups, with the user write groups further divided into cold data groups and hot data groups. Data block lifecycle characteristics are used when grouping user writes, while lifecycle-based characteristics are used when grouping garbage collection write groups.
[0036] As a preferred option, the re-access discriminator consists of multiple cascaded Bloom filters; The implementation of saving the logical block address to the re-access discriminator is as follows: the current logical block address to be saved is inserted into a Bloom filter; when the data inserted into the Bloom filter reaches the threshold, the Bloom filter is cascaded onto the current cascaded Bloom filter. The query is implemented as follows: a query is performed in all Bloom filters in the re-access discriminator. If the number of Bloom filter riders that can find the data exceeds the threshold, it is considered a hit; otherwise, it is considered a miss.
[0037] Specifically, a first-in, first-out (FIFO) strategy is used to limit the number of Bloom filters and to remove the oldest Bloom filters when necessary.
[0038] The proactive degradation placement strategy includes the following steps: predicting the lifespan of data blocks by analyzing their lifecycle and access frequency; building a re-access discriminator to identify data blocks with long lifecycles and predict their future access frequency; and, based on the prediction, downgrading long-lifecycle data blocks from hot groups to cold groups to reduce redundant data migration.
[0039] Data migration optimization during garbage collection includes the following steps: By accurately tracking the lifecycle of data blocks, optimize the candidate segment selection process for garbage collection to ensure that only truly invalid data blocks are migrated; Based on the usage frequency and lifecycle prediction of data blocks, rationally select migration target groups to reduce write amplification; Adjust the migration order of data blocks to optimize the efficiency of data migration and reduce write amplification caused by garbage collection.
[0040] The placement method in this embodiment can adjust the data block allocation strategy in real time according to changes in workload to ensure that the system is always in an optimal performance state; it can adjust the aggregation processing of hot data blocks according to changes in access density to reduce unnecessary migration; and it can dynamically adjust the configuration of the storage system through a multi-objective optimization algorithm to optimize throughput, reduce memory overhead, and maximize resource utilization. The overall architecture of this embodiment is as follows: Figure 2 As shown, to better illustrate the present invention, the following example is given: The experimental environment consisted of four Samsung MZQL2960HCJR-00A07 SSDs, one Intel Xeon Silver 4314 CPU, and 128GB of RAM. The data block size was 4KB, while the default block size of the underlying SSD array was 64KB.
[0041] To demonstrate the efficient update performance of this invention, and to illustrate its effectiveness and superiority, experiments were conducted on various real-world load datasets. For example... Figure 3 and Figure 4 As shown, compared to the five existing mainstream data placement strategies (such as SepGC, MiDA, and WARCIP), this solution reduces write amplification by 21.8%-46.3% and fill traffic by 40%-72.1% in production environments. Under heavy loads, ADAPT maintains high throughput and manageable memory overhead, demonstrating strong adaptability and system performance optimization capabilities.
[0042] Example 2 This application also relates to an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.
[0043] The electronic device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The memory can be used to store computer programs and / or modules. The processor performs various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory, and by accessing data stored in the memory.
[0044] The relevant technical solutions are the same as above, and will not be repeated here.
[0045] Example 3 This application also relates to a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.
[0046] Specifically, the memory may include high-speed random access memory, as well as non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital (SD) cards, flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.
[0047] The relevant technical solutions are the same as above, and will not be repeated here.
[0048] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A log-structured storage data placement method, characterized by, Comprise: S1, according to the access frequency and the hot degree of each data block corresponding to the user write request flow, the data block is classified as hot data block or cold data block, and is grouped to hot data group or cold data group accordingly; S2, the proportion of zero padding data block in each data group in the process of persisting is calculated in real time, when the proportion of zero padding data block of hot data group and cold data group is greater than the preset threshold value, cross-group aggregation is triggered, the cross-group aggregation mode is: the hot data block corresponding to the current user load being persisted in the hot data group is recorded as cross-group data, the cold data block corresponding to the current user load being persisted and the cross-group data constitute aggregated data, if the current aggregated data size is aligned with the large block in the storage array, and / or the aggregated data persistence timer is overdue, the aggregated data is persisted, the temporary storage of the cross-group data is realized;The hot data group persistence processing timer is reset, the newly classified hot data block is continuously received, after the hot data group persistence processing timer is overdue, the last cross-group data in the hot data group and the newly received hot data block after it are persisted together, the delayed persistence of the last cross-group data is realized, after the persistence processing, the last cross-group data is deleted in the aggregated data persistence;When the cross-group aggregation condition is not met, each data group is persisted respectively when the corresponding persistence timer is overdue, and finally the placement management of the log structure storage data is realized.
2. A log-structured data placement method as claimed in claim 1, characterized in that, The last access interval t of the data block is used to represent the access frequency and the hot degree; The implementation mode of classifying the data block is: The last access interval t of the data block is compared with the optimal classification threshold value of the log structure storage system, if t is greater than the optimal classification threshold value, the data block is classified as cold data block, otherwise, the data block is classified as hot data block.
3. A log-structured data placement method as claimed in claim 2, characterized in that, The optimal classification threshold value is dynamically adjusted by the following method: The user write request flow is sampled, a part of write requests are randomly selected from the write request flow at a preset fixed sampling rate, and the data blocks corresponding to each sampled write request are determined. The determined data blocks are respectively simulated on preset M ghost sets: the last access interval of each data block is calculated, and the last access interval and the cold-hot separation threshold of each ghost set are compared If the last access interval is greater than , the data block is divided into a cold data group, otherwise, the data block is divided into a hot group; each ghost set independently runs a garbage collection process and records its write amplification performance; The write amplification performance of all ghost sets is compared, and the cold-hot separation threshold value corresponding to the minimum write amplification is selected as the optimal separation threshold value of the log structure storage system.
4. A log-structured data placement method as claimed in claim 3, characterized in that, The M cold-hot separation threshold values are initialized in an exponential growth relationship; When there is an optimal separation threshold value, the subsequent sampling process adjusts the M cold-hot separation threshold values in a linear relationship;When the cumulative write request amount of the user exceeds 10% of the storage capacity or the write amplification tends to be stable, the cold-hot separation threshold value is reinitialized.
5. The log-structured data placement method of claim 1 wherein, The zero-padding data block proportion The determination manner is that: wherein represents the number of data blocks placed in the data group i represents the number of data blocks placed in the data group i represents the number of data blocks placed in the data group represents the size of each data block, represents the number of data blocks contained in one segment in the log-structured storage system, is the number of times of zero padding in the data group i is the number of times of zero padding in the data group k represents the number of data blocks placed in the data group i represents the number of data blocks placed in the data group 6. The log-structured data placement method of claim 1 wherein, Before S1, the method further comprises: For each user write request corresponding data block, query the pre-constructed re-access discriminator, if hit, the data block is directly migrated to the garbage collection write group corresponding to the life cycle;Otherwise, S1 is executed; The re-access discriminator is constructed by the following method: When migrating the valid data block in the garbage collection process, according to the selected recovery section, if the data block is migrated back to the original garbage collection write group after being re-identified according to the life cycle, the logical block address of the data block is extracted and saved to the re-access discriminator.
7. A log-structured data placement method as claimed in claim 6, characterized in that, The re-access discriminator is composed of a plurality of Bloom filters in cascade; The implementation of saving the logical block address into the re-access discriminator is: inserting the current logical block address to be saved into one Bloom filter; when the data inserted into the Bloom filter reaches a threshold, cascading the Bloom filter to the current cascaded Bloom filter; The implementation of the query is: querying in all the Bloom filters in the re-access discriminator, and if the number of Bloom filters in which the data is queried exceeds the threshold, it is considered as a hit, otherwise, it is considered as a miss. 8.An electronic device comprising a memory and a processor, the memory storing a computer program, wherein, The processor implements the steps of the method of any one of claims 1 to 7 when executing the computer program.
9. A computer-readable storage medium, characterized in that, The computer readable storage medium comprises a stored computer program, wherein the computer program, when executed by a processor, controls the device in which the storage medium is located to perform the steps of the method of any one of claims 1 to 7.