A virtual storage-based mobile solid state disk management method

Abstract

This invention discloses a method for managing a portable solid-state drive (SSD) based on virtual storage, comprising the following steps: Step 1: Reading the logical address arrangement and operating status of the portable SSD, and performing continuity detection and correction processing; Step 2: Performing address location, time window division, and segment type extraction on the currently mounted file access records; Step 3: Performing time-series prediction using an improved FEDformer model to obtain access migration results; Step 4: Performing migration rearrangement and segment filtering on the virtual storage organization table based on the access migration results, generating a rearrangement mapping table; Step 5: Performing sequential scanning and splitting processing on actual read / write requests based on the rearrangement mapping table, and generating an execution scheduling table; Step 6: Recording and encapsulating recovery status units based on the execution scheduling table, and generating a breakpoint recovery table. This invention achieves predictive rearrangement and breakpoint recovery for portable SSDs through an improved FEDformer model.

Description

Technical Field

[0001] This invention relates to the field of mobile storage management technology, and in particular to a method for managing mobile solid-state drives based on virtual storage. Background Technology

[0002] With the increasing demand for mobile office, cross-device collaborative processing, and portable data storage, portable solid-state drives (SSDs) are widely used in scenarios such as engineering file transfer, image and video processing, program development, database copying, and data exchange between multiple hosts due to their fast read and write speeds, high capacity, and portability. However, in practical applications, most existing portable SSDs still rely on fixed logical address mapping and conventional read / write scheduling to access data, typically depending solely on the underlying file system or controller for address management, cache allocation, and exception recovery. When portable SSDs frequently connect to different hosts, or face complex access scenarios involving continuous read / write, alternating calls, and partial overwriting, existing management methods often only passively respond to the current request, making it difficult to proactively handle subsequent data migration and segment adjustments based on the access process within the session.

[0003] In existing technologies, on the one hand, the logical address arrangement state and the running state retained before the last unloading are usually handled separately. There is a lack of a unified continuity detection and correction mechanism for situations such as boundary overlap, address jumps, and unclosed write-backs, which easily leads to unclear organizational segment divisions and unstable write-back connections. On the other hand, existing solutions mostly handle file access processes at the single request level, lacking a process for merging, sorting, and extracting segment types based on organizational segment positions and time window relationships. Therefore, it is difficult to form a session access sequence that can characterize the current session access pattern. Furthermore, existing mobile SSD management methods typically lack the ability to combine time-series prediction models to predict the access migration direction for the next period. The prediction results cannot be used for organizational segment migration rearrangement, read / write request order shaping, and breakpoint recovery linkage after abnormal disconnections. This results in numerous read / write request jumps, discontinuous migration orchestration, and reliance on simple rewrites during recovery, leading to insufficient overall management continuity and targeting.

[0004] Therefore, how to provide a mobile solid-state drive management method based on virtual storage is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a mobile solid-state drive (SSD) management method based on virtual storage. This invention achieves access migration prediction, organizational segment rearrangement, read / write order shaping, and abnormal breakpoint recovery for mobile SSDs under complex sessions by constructing a virtual storage organization table, session access sequence, and an improved FEDformer prediction link, thereby improving storage organization continuity, scheduling order, and recovery integrity.

[0006] A mobile solid-state drive management method based on virtual storage according to an embodiment of the present invention includes the following steps: Step 1: Read the current logical address arrangement state of the portable solid-state drive and the running state retained before the last unloading, and perform continuity detection and correction to form a virtual storage organization table; Step 2: Based on the virtual storage organization table, locate the address, divide the time window, and extract the segment type of the currently mounted file access records to form a session access sequence; Step 3: Input the session access sequence into the improved FEDformer model, and perform time series prediction through the frequency domain decomposition module, trend coding module, fusion inference module and migration decoding module to obtain the access migration results for the next time period; Step 4: Based on the access migration results, perform migration rearrangement on the virtual storage organization table, and generate a rearrangement mapping table by filtering by segment; Step 5: Based on the rearranged mapping table, perform a sequential scan of the actual read and write requests in the current session to form a continuous read and write segment queue, split the requests to form a segmented execution queue, and generate an execution schedule table; Step 6: Encapsulate the recovery status unit according to the order of issuance recorded in the execution schedule table, and generate the breakpoint recovery table.

[0007] Optionally, step one specifically includes: After the external solid-state drive is connected to the host, read the correspondence between the current logical address and the physical storage block, and the running state retained before the last unloading; The starting position, ending position, and order of the physical storage block corresponding to the current logical address segment are superimposed with the running status according to the corresponding logical address segment to form a status superposition queue. Perform continuity detection on adjacent logical address segments in the state overlay queue. The continuity detection is to mark the area where the termination position intersects and overlaps with the starting position of the next logical address segment as the boundary overlap position. The regions where the physical storage block numbers are not continuous between adjacent logical address segments are marked as address transition locations, and the regions where there are unwritten data blocks after the written boundary are marked as unclosed write-back locations, forming an abnormal location marking queue. Based on the abnormal location marker queue, a correction process is performed on each abnormal location. The correction process is to retain the termination boundary of the preceding logical address segment and truncate the overlapping part of the following logical address segment at the boundary overlap location. Extract the empty physical storage block between the address segment before the address jump and the address segment after the address jump position and perform padding and rearrangement. For unclosed write-back positions, call the tail write record retained before the last unloading to perform padding and closure, forming a corrected address queue. The logical address segments in the corrected address queue are reordered according to their physical location. Logical address segments that are physically continuous and do not have overlapping boundaries, address jumps, or unclosed write-backs are merged into the same organization segment. Then, the segment start address, segment end address, and internal arrangement order are written to each organization segment in sequence to form a virtual memory organization table.

[0008] Optionally, step two specifically includes: Read the logical address and organization segment of each file currently mounted in the virtual storage organization table, and record each read action, write action and call switch action in the order in which the file access occurs, forming an access record queue; Perform address location processing and time window division on each file access record in the access record queue to obtain an access cluster set; Segment type extraction is performed on each access cluster in the access cluster set. The segment type extraction is to extract consecutive file access records along the organizational segment order within the same access cluster into consecutive access segments. Extract file access records that switch back and forth between two or more files within the same access cluster into cross access segments; extract file access records between the end of the previous write operation and the start of the next write-back operation within the same access cluster into write-back interval segments, forming a set of access segments; According to the distribution order of each access cluster in the virtual memory organization table, the consecutive access segments, intersecting access segments, and write-back interval segments in the access segment set are rearranged. The rearrangement is to place the access segments located in the preceding organization segment before the access segments located in the following organization segment, and then sort the access segments located in the same organization segment according to the actual occurrence time to form the segment sorting result. Based on the segment sorting results, the sorted consecutive access segments, cross access segments, and write-back interval segments are concatenated in chronological order, and segment switching markers and time connection markers are written between adjacent access segments to form a session access sequence.

[0009] Optionally, the step of performing address location processing and time window division on each file access record in the access record queue to obtain an access cluster set is as follows: Each file access record is mapped to an organization segment in the virtual storage organization table. When multiple file access records correspond to the same or adjacent organization segments, the current multiple file access records are marked as segment adjacency access records. When multiple file access records correspond to different and non-adjacent organizational sections, the current multiple file access records are marked as section jump access records, forming a section location result table; Based on the segment location result table, the time window is divided for each file access record. File access records that appear consecutively within a preset time length are divided into the same time window, and continuous read actions, continuous rewrite actions, and alternating call actions are identified within each time window. When multiple file access records within the same time window simultaneously meet the conditions of segment adjacency and action continuity, the current multiple files are merged into the same access cluster to form an access cluster set.

[0010] Optionally, the improved FEDformer model is specifically as follows: The session access sequence is input into the frequency domain decomposition module, and frequency domain splitting is performed to obtain the decomposed sequence group. The decomposed sequence group is input into the trend encoding module. Multiple sliding windows of different lengths are set for the periodic component sequence and the burst component sequence, and local sequence segments are extracted along the time axis segment by segment using each sliding window. For two adjacent local sequence segments in the same component sequence, perform head-to-tail alignment, calculate the difference in the first value of the segment, the difference in the last value of the segment, and the change in the slope of the middle segment, and then concatenate multiple consecutive local sequence segments according to the window advancement order to obtain the segment change chain at each time scale; Arrange the segments of change at each time scale according to their starting point to form a trend trajectory group; Input the trend trajectory group into the fusion and inference module, align the periodic trajectory and sudden trajectory in the trend trajectory group according to the same time position, and use the periodic trajectory as the main sequence to superimpose the sudden trajectory segment by segment onto the corresponding time position to obtain the combined trajectory sequence. The combined trajectory sequence is recursively pushed forward according to the preset prediction length, and prediction segments at different time positions are generated one by one. The prediction segments are then spliced ​​together in the order of generation to form a prediction access sequence. In the predicted access sequence, the direction of access intensity change, the direction of access position advancement, and the direction of independent access enhancement are scanned segment by segment, and the scan results are merged to form a migration determination group; The migration determination group is input into the migration decoding module. Each determination segment in the migration determination group is expanded according to the corresponding file. Files that are determined to advance the access position and whose access intensity increases beyond the preset threshold are marked for forward movement. Files that have completed the forward movement mark are written into the forward movement queue in sequence. Files that are determined to have been moved backwards in access positions and whose access intensity has decreased by more than a preset threshold are marked as moved backwards, and the files marked as moved backwards are written into the queue to be moved backwards in sequence. Files that are determined to have an increase in independent access counts exceeding a preset threshold and a decrease in cross-segment switching counts exceeding a preset threshold are marked for isolation, and the files that have been marked for isolation are written into the isolation queue in sequence. The queues to be moved forward, to be moved backward, and to be isolated are deduplicated, sorted, and output to form the access migration results.

[0011] Optionally, the step of inputting the session access sequence to the frequency domain decomposition module and performing frequency domain splitting processing to obtain a decomposed sequence group specifically involves: The session access sequence is divided into multiple continuous subsequences with a fixed step size; For each continuous subsequence, the corresponding spectrum expansion result is obtained by performing a fast Fourier transform. In each spectrum expansion result, each frequency component is scanned in order from low to high frequency position. Frequency components that appear repeatedly in adjacent subsequences are retained as stable components, and frequency components that appear only in local subsequences are extracted as abrupt components. Inverse transformations are performed on the stable components and the mutation components respectively. The stable components after inverse transformation are then spliced ​​together in the original segmentation order to form a periodic component sequence. The mutation components after inverse transformation are then spliced ​​together in the original segmentation order to form a burst component sequence. The periodic component sequences and burst component sequences are aligned according to their time positions to form a decomposed sequence group.

[0012] Optionally, step four specifically includes: For each file in the forward set, backward set, and isolated set of access migration results, read the current logical address and the organization segment it belongs to, and record the queue position according to the original arrangement order of each file in the virtual storage organization table to form a migration orchestration queue. Based on the migration orchestration queue, each file in the forward queue is filtered for forward shift segments to form a forward shift placement table; Based on the migration orchestration queue, each file in the move-back queue is filtered for move-back segments to form a move-back placement table; Based on the migration orchestration queue, each file in the isolation queue is stripped and transferred to form an isolation placement table; Each file in the forward placement table, backward placement table, and isolated placement table is read from its original address segment and written to its corresponding forward placement address segment, backward placement address segment, and corresponding independent organization segment address segment. After each file is written, the address position occupied by the original address segment of the current file is released to form the actual placement table. Based on the actual location table, the correspondence between the file logical address and the organizational segment is rewritten according to the actual location relationship after file migration. The new logical address start point, new logical address end point and new organizational segment of each file after migration are written into the mapping record area. The records in the mapping record area are reordered according to the organizational segment arrangement order and the address order of each file in the segment to form a rearranged mapping table.

[0013] Optionally, step five specifically includes: Read the file logical address range and organizational segment position corresponding to each read / write request in the current session, map each read / write request to the corresponding record in the rearranged mapping table in the order of arrival, and write the request start address, request end address, segment number and request type to each read / write request to form a request mapping queue. The read and write requests in the request mapping queue are scanned sequentially according to the organization segment number. The sequential scan is to concatenate discrete read and write requests with the same segment number and an address interval not exceeding a preset span in sequence, perform segment filling registration on the gap between the end address of the previous read and write request and the start address of the next read and write request, and merge the concatenated request fragments in ascending order of address to form a continuous read and write segment queue. For cross-segment read / write requests in the continuous read / write segment queue, a splitting process is performed. The splitting process involves cutting each cross-segment read / write request into segments according to the boundaries of the covered organizational segments, resulting in multiple sub-request segments corresponding to different organizational segments. Arrange the sub-request segments belonging to the same organizational segment in the original order of request arrival, and write the arranged sub-request segments one by one into the execution sub-queue of the corresponding segment to form a segmented execution queue. Based on the request mapping queue, controlled transfer processing is performed on access requests that fall into an independent organization segment to obtain a controlled access queue; Based on the continuous read / write segment queue, the segmented execution queue, and the controlled access queue, a unified distribution sorting is performed according to the organizational segment arrangement order in the rearranged mapping table to form a distribution order queue. Each consecutive read / write segment, execution subqueue, and controlled access request in the distribution sequence queue is sequentially written into the preset scheduling record area. When writing each item, the corresponding distribution sequence number, the organization segment to which it belongs, the execution start address, the execution end address, and the preceding and following connection positions are written synchronously. All records in the scheduling record area are ordered and arranged according to the distribution sequence number to form an execution scheduling table.

[0014] Optionally, step six specifically includes: Data reading, data writing, and segmented write-back are executed according to the order of issuance recorded in the execution schedule table. After each round of continuous read / write segment execution or segmented execution queue write-back is completed, the corresponding latest address position, completed write boundary, and incomplete write-back position are recorded and encapsulated into a recovery state unit. When an abnormal disconnection or abnormal unloading is detected, the current writing operation that has not yet been completed is stopped, the incomplete write-back position in the recovery status unit is extracted, the tail writing is performed on the data block corresponding to the incomplete write-back position, the boundary closure is performed on the data segment after the completed write boundary that has not yet been closed, and the writing result and the closure result are written to the recovery retention area to form the breakpoint recovery table. After the external SSD is connected to the host again, the order of data blocks corresponding to the incomplete write-back positions is restored according to the breakpoint recovery table.

[0015] The beneficial effects of this invention are: This invention addresses common problems encountered by mobile solid-state drives (SSDs) in multi-host scenarios, such as discrete logical address organization, lack of continuous modeling of access processes, delayed data migration and adjustment, frequent read / write scheduling jumps, and poor continuity of recovery after abnormal unloading. It constructs a complete processing chain from address organization correction to access prediction, migration scheduling, and breakpoint recovery. By reading the current logical address arrangement and the operating state retained before the last unloading, and performing continuity detection and correction, it can normalize abnormal locations such as boundary overlaps, address jumps, and unclosed write-backs during the organization phase, thereby improving the stability of the virtual storage organization table and the consistency of subsequent processing. By performing address location, time window division, and segment type extraction on the currently mounted file access records, it can transform discrete read, write, and call switching actions into a session access sequence with temporal relationships, enabling a structured expression of the actual access patterns in the current session.

[0016] By inputting the session access sequence into the improved FEDformer model and performing hierarchical processing using frequency domain decomposition, trend encoding, fusion inference, and migration decoding modules, the access migration direction for the next time period can be predicted in a targeted manner, providing a clear basis for the division of files to be moved forward, files to be moved backward, and files to be isolated. After performing migration rearrangement based on the access migration results, the placement relationship of files in the organizational segments can be made more consistent with the access trend of the current session and the next time period, reducing disordered jumps and invalid migrations. Furthermore, by combining the rearrangement mapping table with sequential scanning, splitting, and unified sorting of actual read and write requests, the originally discrete or cross-segment read and write processes can be reorganized into a more ordered execution scheduling process. Finally, the recovery state unit is encapsulated based on the execution scheduling table, and a breakpoint recovery table is generated, providing a clear basis for data rewriting, boundary closure, and order restoration after abnormal disconnection or abnormal unloading. Therefore, this invention can enhance the organizational continuity, migration targeting, scheduling order, and recovery integrity of portable solid-state drives under complex session conditions. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a mobile solid-state drive management method based on virtual storage proposed in this invention; Figure 2This is a schematic diagram illustrating the session access sequence generation steps of a mobile solid-state drive management method based on virtual storage proposed in this invention. Figure 3 This is a flowchart of the improved FEDformer model processing procedure for a mobile solid-state drive management method based on virtual storage proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figures 1-3 A method for managing portable solid-state drives based on virtual storage includes the following steps: Step 1: Read the current logical address arrangement state of the portable solid-state drive and the running state retained before the last unloading, and perform continuity detection and correction to form a virtual storage organization table; Step 2: Based on the virtual storage organization table, locate the address, divide the time window, and extract the segment type of the currently mounted file access records to form a session access sequence; Step 3: Input the session access sequence into the improved FEDformer model, and perform time series prediction through the frequency domain decomposition module, trend coding module, fusion inference module and migration decoding module to obtain the access migration results for the next time period; Step 4: Based on the access migration results, perform migration rearrangement on the virtual storage organization table, and generate a rearrangement mapping table by filtering by segment; Step 5: Based on the rearranged mapping table, perform a sequential scan of the actual read and write requests in the current session to form a continuous read and write segment queue, split the requests to form a segmented execution queue, and generate an execution schedule table; Step 6: Encapsulate the recovery status unit according to the order of issuance recorded in the execution schedule table, and generate the breakpoint recovery table.

[0020] This step addresses the problems of address organization disorder, difficulty in utilizing access patterns, delayed migration adjustments, discrete read / write requests, and poor recovery continuity after abnormal unloading of portable SSDs in multi-host scenarios. It constructs an integrated processing mechanism encompassing virtual storage organization, access sequence extraction, timing prediction, migration rearrangement, execution scheduling, and breakpoint recovery. By continuously detecting and correcting the logical address arrangement state and historical operating state, the stability of the organized segment division can be improved. By temporally merging file access records and combining them with an improved FEDformer model for access migration prediction, the targeted nature of subsequent segment adjustments can be enhanced. Through the coordinated processing of remapping mapping and execution scheduling, disordered access caused by read / write jumps can be reduced. By combining recovery state units and breakpoint recovery tables, the data continuation capability after abnormal disconnection can be improved, thereby enhancing the continuity, orderliness, and recovery reliability of the portable SSD management process.

[0021] In this embodiment, step one specifically includes: After the external SSD is connected to the host, it responds to the mount command of the external SSD connected to the host, reads the correspondence between the current logical address and the physical storage block, and the running state retained before the last unmount; The operational status includes the completed write boundary, the incomplete write-back position, the address migration termination position, and the organizational segment closure status. The starting position, ending position, and order of the physical storage block corresponding to the current logical address segment are superimposed with the running status according to the corresponding logical address segment to form a status superposition queue. Perform continuity detection on adjacent logical address segments in the state overlay queue. The continuity detection is to mark the area where the termination position intersects and overlaps with the starting position of the next logical address segment as the boundary overlap position. The regions where the physical storage block numbers are not continuous between adjacent logical address segments are marked as address transition locations, and the regions where there are unwritten data blocks after the written boundary are marked as unclosed write-back locations, forming an abnormal location marking queue. Based on the abnormal location marker queue, a correction process is performed on each abnormal location. The correction process involves retaining the termination boundary of the preceding logical address segment and truncating the overlapping portion of the following logical address segment at the boundary overlap location. Extract the empty physical storage block between the address segment before the address jump and the address segment after the address jump position and perform padding and rearrangement. For unclosed write-back positions, call the tail write record retained before the last unloading to perform padding and closure, forming a corrected address queue. The logical address segments in the corrected address queue are reordered according to their physical location. Logical address segments that are physically continuous and do not have overlapping boundaries, address jumps, or unclosed write-backs are merged into the same organization segment. Then, the segment start address, segment end address, and internal arrangement order are written to each organization segment in sequence to form a virtual memory organization table.

[0022] This step involves overlaying the current logical address mapping with the operational state retained before the last unmounting after the external SSD is connected to the host. It also performs unified detection and correction of abnormal locations such as overlapping boundaries, address jumps, and incomplete write-backs. This allows for timely elimination of structural misalignments caused by abnormal unmounting, interrupted address migration, or incomplete write-backs during the initial data organization phase, ensuring that subsequent virtual storage organization is built on a more stable and continuous address foundation. Furthermore, by reordering and merging the corrected logical address segments according to their physical locations to form organizational segments, the address distribution within the external SSD becomes more regular, reducing disordered jumps and redundant switching during subsequent access. Compared to existing technologies that rely solely on the underlying file system to passively handle abnormal states, this invention proactively repairs address continuity and rebuilds organizational segments during the initial mounting phase, thereby improving the accuracy of virtual storage organization, the consistency of subsequent migration orchestration, and the reliability of data connection in abnormal scenarios.

[0023] In this embodiment, step two specifically includes: Read the logical address and organization segment of each file currently mounted in the virtual storage organization table, and record each read action, write action and call switch action in the order in which the file access occurs, forming an access record queue; Perform address location processing on each file access record in the access record queue, wherein the address location processing is to map each file access record to an organization segment in the virtual storage organization table; When multiple file access records correspond to the same or adjacent organizational segments, these multiple file access records are marked as segment contiguous access records. When multiple file access records correspond to different and non-adjacent organizational sections, the current multiple file access records are marked as section jump access records, forming a section location result table; Based on the segment location result table, the time window is divided for each file access record. File access records that appear consecutively within a preset time length are divided into the same time window, and continuous read actions, continuous rewrite actions, and alternating call actions are identified within each time window. When multiple file access records within the same time window simultaneously meet the conditions of segment adjacency and action continuity, the current multiple files are merged into the same access cluster to form an access cluster set; Segment type extraction is performed on each access cluster in the access cluster set. The segment type extraction is to extract consecutive file access records along the organizational segment order within the same access cluster into consecutive access segments. Extract file access records that switch back and forth between two or more files within the same access cluster into cross-access segments; Extract the file access records between the end of the previous write-back action and the start of the next write-back action within the same access cluster into write-back interval segments, forming an access segment set; According to the distribution order of each access cluster in the virtual memory organization table, the consecutive access segments, intersecting access segments, and write-back interval segments in the access segment set are rearranged. The rearrangement is to place the access segments located in the preceding organization segment before the access segments located in the following organization segment, and then sort the access segments located in the same organization segment according to the actual occurrence time to form the segment sorting result. The segment sorting result is called, and the sorted consecutive access segments, cross access segments, and write-back interval segments are concatenated in chronological order. Segment switching markers and time connection markers are written between adjacent access segments to form a session access sequence.

[0024] This step maps the currently mounted file access process to a virtual storage organization table and, by combining the organizational segment position relationships, time window relationships, and access action continuity relationships, merges and organizes discrete read, write, and call switching records. This transforms the originally scattered file access behaviors into a session access sequence with a clear temporal structure. By distinguishing between segment adjacency access records and segment jump access records, and further extracting continuous access segments, cross-access segments, and write-back interval segments, the actual access patterns and segment switching rules in the current session can be more accurately reflected, providing a more targeted input basis for subsequent access migration prediction. Compared to existing technologies that only process single access requests in isolation, this invention enhances the ability to express the continuity, correlation, and stages of access behavior, thereby improving the matching degree of subsequent model prediction, segment migration rearrangement, and execution scheduling.

[0025] In this embodiment, the improved FEDformer model is specifically as follows: The session access sequence is input into the frequency domain decomposition module, and frequency domain splitting is performed to obtain the decomposed sequence group. The decomposed sequence group is input into the trend encoding module. Multiple sliding windows of different lengths are set for the periodic component sequence and the burst component sequence, and local sequence segments are extracted along the time axis segment by segment using each sliding window. For two adjacent local sequence segments in the same component sequence, perform head-to-tail alignment, calculate the difference in the first value of the segment, the difference in the last value of the segment, and the change in the slope of the middle segment, and then concatenate multiple consecutive local sequence segments according to the window advancement order to obtain the segment change chain at each time scale; Arrange the segments of change at each time scale according to their starting point to form a trend trajectory group; Input the trend trajectory group into the fusion and inference module, align the periodic trajectory and sudden trajectory in the trend trajectory group according to the same time position, and use the periodic trajectory as the main sequence to superimpose the sudden trajectory segment by segment onto the corresponding time position to obtain the combined trajectory sequence. The combined trajectory sequence is recursively pushed forward according to the preset prediction length, and prediction segments at different time positions are generated one by one. The prediction segments are then spliced ​​together in the order of generation to form a prediction access sequence. In the predicted access sequence, the direction of access intensity change, the direction of access position advancement, and the direction of independent access enhancement are scanned segment by segment, and the scan results are merged to form a migration determination group; The migration determination group is input into the migration decoding module. Each determination segment in the migration determination group is expanded according to the corresponding file. Files that are determined to advance the access position and whose access intensity increases beyond the preset threshold are marked for forward movement. Files that have completed the forward movement mark are written into the forward movement queue in sequence. Files that are determined to have been moved backwards in access positions and whose access intensity has decreased by more than a preset threshold are marked as moved backwards, and the files marked as moved backwards are written into the queue to be moved backwards in sequence. Files that are determined to have an increase in independent access counts exceeding a preset threshold and a decrease in cross-segment switching counts exceeding a preset threshold are marked for isolation, and the files that have been marked for isolation are written into the isolation queue in sequence. The queues to be moved forward, to be moved backward, and to be isolated are deduplicated, sorted, and output to form the access migration results.

[0026] The improved FEDformer model proposed in this step shares similarities with the traditional FEDformer model in that both are processing models for time series forecasting, and both generally follow the basic processing approach of "first decomposing the input sequence, then encoding and modeling different components, and finally completing the prediction output." Both retain the core idea of ​​FEDformer for trend analysis of long-term time series data, both emphasize the separation of stable and fluctuating components in the sequence, and both reduce interference when directly predicting from the original sequence through component-level modeling. Furthermore, both models exhibit a serial processing logic consisting of front-end decomposition, intermediate encoding, and back-end prediction. The inputs are time series with temporal relationships, and the outputs are prediction results for the next or subsequent time periods. From a functional perspective, both the traditional FEDformer and the improved FEDformer model in this step serve to characterize trends, extrapolate changes, and predict subsequent states of complex time series. Neither is a simple static classification model, but rather extracts the continuity, fluctuation, and evolution patterns in the sequence through temporal context. Therefore, they are consistent in their basic model framework, time series decomposition ideas, and overall direction of predicting future sequence changes.

[0027] The difference lies in the fact that the improved FEDformer model in this step does not remain at the level of traditional FEDformer's numerical prediction of general time series, but rather undergoes a structured transformation oriented towards business actions for session access sequences in mobile SSD management scenarios. Traditional FEDformer typically outputs predicted values ​​for future time points or future intervals, focusing on improving time series prediction accuracy. The improved model in this step, after frequency domain decomposition, introduces a trend encoding module to set multiple sliding windows of different lengths for both periodic and burst component sequences. It constructs segment change chains through the difference between the first and last values ​​and changes in the middle slope, enabling the model to not only focus on numerical changes but also on the trajectory of access behavior at different time scales. Furthermore, this step adds a fusion inference module, aligning and superimposing periodic and burst trajectories at the same time position, not directly obtaining a single predicted value, but forming a predicted access sequence and migration decision group. More importantly, this step includes a migration decoding module that expands the prediction results to the specific file level, forming queues to be moved forward, queues to be moved backward, and queues to be isolated according to three actions: moving forward, moving backward, and isolating. This allows the model output to directly serve the subsequent migration and rearrangement of the virtual storage organization table. Therefore, this improved model has been transformed from a general time-series prediction structure into a dedicated prediction structure for storage scheduling decisions.

[0028] The beneficial effect of the improvements lies in combining the time-series decomposition capabilities of traditional FEDformer with the determination of file access behavior, organizational segment changes, and migration actions in mobile SSD management. This transforms the model output from abstract predictions that are difficult to apply directly into access migration results that can directly drive forward, backward, and isolation operations. Because this step performs sliding window truncation, segment alignment, and change chain concatenation on periodic and burst components respectively, it can more precisely characterize the differences between stable access patterns and local burst access patterns in the session access sequence, improving the ability to identify the direction of access changes in the next time period. By uniformly aligning and overlaying different trajectories according to their time positions through the fusion and inference module, the prediction results can more closely approximate the complex change states in the actual access process, avoiding migration bias caused by single-trend judgments. Furthermore, the migration decoding module refines the prediction results to the specific file level, providing clear objects and directions for subsequent migration rearrangement, thereby enhancing the targeted nature of virtual storage organization adjustments. Overall, this improvement not only enhances the model's ability to express complex access sequences, but also increases the directness of the conversion of prediction results into storage management actions, enabling subsequent reordering mapping, read / write scheduling, and breakpoint recovery to be based on more forward-looking access judgments.

[0029] In this embodiment, the step of inputting the session access sequence to the frequency domain decomposition module and performing frequency domain splitting to obtain the decomposed sequence group specifically involves: The session access sequence is divided into multiple continuous subsequences with a fixed step size; For each continuous subsequence, the corresponding spectrum expansion result is obtained by performing a fast Fourier transform. In each spectrum expansion result, each frequency component is scanned in order from low to high frequency position. Frequency components that appear repeatedly in adjacent subsequences are retained as stable components, and frequency components that appear only in local subsequences are extracted as abrupt components. Inverse transformations are performed on the stable components and the mutation components respectively. The stable components after inverse transformation are then spliced ​​together in the original segmentation order to form a periodic component sequence. The mutation components after inverse transformation are then spliced ​​together in the original segmentation order to form a burst component sequence. The periodic component sequences and burst component sequences are aligned according to their time positions to form a decomposed sequence group.

[0030] This step involves sequentially inputting the session access sequence into the frequency domain decomposition module, trend encoding module, fusion and deduction module, and migration decoding module. This process performs layered decomposition and continuous deduction of access changes within the current session of the portable solid-state drive (SSD). It extracts the previously mixed periodic access fluctuations and localized burst access fluctuations, then unifies and fuses these by combining fragment change chains at different time scales. This ensures that subsequent access trend judgments are no longer based on a single moment or single request, but rather on continuous temporal changes. Furthermore, by scanning the direction of access intensity change, access position advancement, and independent access enhancement in the predicted access sequence segment by segment, and forming queues to be moved forward, moved backward, and isolated respectively, a clear and executable migration basis is provided for subsequent virtual storage organization rearrangement. Compared to existing technologies that rely on static statistical results or simple access counts for file adjustment, this invention enhances the ability to characterize access migration trends in the next time period, improves the targeting of forward, backward, and isolation judgments, and makes subsequent segment migration, read / write scheduling, and recovery processing more closely reflect real access changes.

[0031] In this embodiment, step four specifically includes: For each file in the forward set, backward set, and isolated set of access migration results, read the current logical address and the organization segment it belongs to, and record the queue position according to the original arrangement order of each file in the virtual storage organization table to form a migration orchestration queue. According to the migration orchestration queue, each file in the forward queue is filtered for forward segments. The forward segment filtering is to scan the front organizational segments one by one along the organizational segment arrangement direction in the virtual storage organizational table, remove organizational segments with non-contiguous addresses and organizational segments with more than a preset number of segment switching records, and merge the remaining organizational segments into the forward candidate segment list in turn. According to the order of the files in the forward queue, each file is loaded into a contiguous free address segment in the forward candidate segment column to form a forward placement table; According to the migration orchestration queue, each file in the move queue is subjected to move segment filtering. The move segment filtering is to scan the rear organizational segments segment by segment along the arrangement direction of the organizational segments in the virtual storage organization table, and extract the organizational segments located after the original organizational segments of the current file as move candidate segments in turn. According to the order of the files in the shift queue, each file is written to the free address range in the shift candidate segment one by one, and the remaining address range of the corresponding shift candidate segment is updated after each write to form the shift placement table; According to the migration orchestration queue, each file in the isolation queue is stripped and transferred. The stripping and transfer involves first reading the address range of the independent organization segment from the virtual storage organization table, and then removing each file in the isolation queue from the original organization segment according to the queue order and writing them one by one into the free address segment of the independent organization segment. After each file is written, the address occupancy relationship between the current file and the original organizational segment is severed, and the new location of the current file in the independent organizational segment is recorded to form an isolation location table; Each file in the forward placement table, backward placement table, and isolated placement table is read from its original address segment and written to its corresponding forward placement address segment, backward placement address segment, and corresponding independent organization segment address segment. After each file is written, the address position occupied by the original address segment of the current file is released to form the actual placement table. Based on the actual location table, the correspondence between the file logical address and the organizational segment is rewritten according to the actual location relationship after file migration. The new logical address start point, new logical address end point and new organizational segment of each file after migration are written into the mapping record area. The records in the mapping record area are reordered according to the organizational segment arrangement order and the address order of each file in the segment to form a rearranged mapping table.

[0032] This step adjusts the file's location within the virtual storage organization based on the access migration results. This ensures that moved files, relocated files, and isolated files are placed in organizational segments more suitable for their subsequent access states. After migration, the correspondence between file logical addresses and organizational segments is rewritten, allowing the virtual storage organization table to be dynamically updated as access changes occur. Compared to existing static storage layouts or simple position adjustments, this invention reduces invalid segment switching and repeated jumps, improves the organizational order and address mapping accuracy after file migration, and provides a more stable foundation for subsequent read / write request shaping and execution scheduling.

[0033] In this embodiment, step five specifically includes: Read the file logical address range and organizational segment position corresponding to each read / write request in the current session, map each read / write request to the corresponding record in the rearranged mapping table in the order of arrival, and write the request start address, request end address, segment number and request type to each read / write request to form a request mapping queue. The read and write requests in the request mapping queue are scanned sequentially according to the organization segment number. The sequential scan is to concatenate discrete read and write requests with the same segment number and an address interval not exceeding a preset span in sequence, perform segment filling registration on the gap between the end address of the previous read and write request and the start address of the next read and write request, and merge the concatenated request fragments in ascending order of address to form a continuous read and write segment queue. For cross-segment read / write requests in the continuous read / write segment queue, a splitting process is performed. The splitting process involves cutting each cross-segment read / write request into segments according to the boundaries of the covered organizational segments, resulting in multiple sub-request segments corresponding to different organizational segments. Arrange the sub-request segments belonging to the same organizational segment in the original order of request arrival, and write the arranged sub-request segments one by one into the execution sub-queue of the corresponding segment to form a segmented execution queue. According to the request mapping queue, the access requests that fall into the independent organization segment are subjected to controlled transfer processing to obtain a controlled access queue. The controlled transfer processing is to filter out the access requests corresponding to the independent organization segment with the segment number from the request mapping queue, write the filtered access requests into a preset controlled entry cache in sequence according to the request arrival time, and attach an access order mark and a segment lock mark to each access request written into the controlled entry cache. The controlled access queue is obtained by transferring the requests into the controlled access queue in the order of writing. Based on the continuous read / write segment queue, the segmented execution queue, and the controlled access queue, a unified distribution sorting is performed according to the organizational segment arrangement order in the rearranged mapping table to form a distribution order queue. The unified distribution sorting involves sorting each continuous read / write segment in the continuous read / write segment queue according to its position in the organizational segment it belongs to, inserting each execution sub-queue in the segmented execution queue into its corresponding sorting position according to its position in the organizational segment it belongs to, and writing each access request in the controlled access queue into the sorted tail controlled interval according to its position in the independent organizational segment, thus forming a distribution order queue. Each consecutive read / write segment, execution subqueue, and controlled access request in the distribution sequence queue is sequentially written into the preset scheduling record area. When writing each item, the corresponding distribution sequence number, the organization segment to which it belongs, the execution start address, the execution end address, and the preceding and following connection positions are written synchronously. All records in the scheduling record area are ordered and arranged according to the distribution sequence number to form an execution scheduling table.

[0034] This step maps read / write requests in the current session to a rearrangement mapping table and performs sequential scanning, segment registration, cross-segment splitting, and controlled transfer processing on discrete requests. This reorganizes previously scattered, abrupt, or restricted access requests into a continuous read / write segment queue, a segmented execution queue, and a controlled access queue. Then, based on the order of the organizational segments, a unified distribution and sorting process is completed, forming an execution schedule table with a clear execution priority. Compared to existing technologies that directly process read / write requests in arrival order, this invention reduces disordered access caused by cross-segment jumps, improves the continuity of request execution within the same segment, and provides access requests in independent organizational segments with a clearer controlled scheduling path, thereby enhancing the scheduling orderliness and execution continuity of the mobile solid-state drive read / write process.

[0035] In this embodiment, step six specifically includes: Data reading, data writing, and segmented write-back are executed according to the order of issuance recorded in the execution schedule table. After each round of continuous read / write segment execution or segmented execution queue write-back is completed, the corresponding latest address position, completed write boundary, and incomplete write-back position are recorded and encapsulated into a recovery state unit. When an abnormal disconnection or abnormal unloading is detected, the current writing operation that has not yet been completed is stopped, the incomplete write-back position in the recovery status unit is extracted, the tail writing is performed on the data block corresponding to the incomplete write-back position, the boundary closure is performed on the data segment after the completed write boundary that has not yet been closed, and the writing result and the closure result are written to the recovery retention area to form the breakpoint recovery table. After the external SSD is connected to the host again, the order of data blocks corresponding to the incomplete write-back positions is restored according to the breakpoint recovery table.

[0036] This step executes data reading, data writing, and segmented write-back according to the order recorded in the execution schedule table, continuously encapsulating recovery status units during execution. This allows for the synchronous preservation of the write progress, address advancement position, and incomplete write-back positions in the current session. In the event of an abnormal disconnection or abnormal unloading, the recovery status units can be used to perform tail-end writing on incomplete write-back positions and boundary closure on unclosed data segments, forming a breakpoint recovery table for sequential recovery upon subsequent access. Compared to existing technologies that rely on simple retries or conventional log recovery, this invention improves the accuracy of data continuation after abnormal interruptions, the continuity of the recovery process, and the integrity of the write-back status preservation.

[0037] Example 1: To verify the feasibility of this invention in practice, it was applied to the post-production workstation area, material processing area, and data review area of ​​a digital content production center. In this scenario, the same portable solid-state drive needs to be repeatedly switched between a high-performance graphics workstation, a portable editing terminal, and a data archiving host. The hard drive simultaneously stores original video materials, proxy files, project cache files, subtitle files, audio mixing files, and periodically exported files. Due to the significant differences in access methods among the different hosts, the graphics workstation performs more continuous reading and large file rewriting, the portable terminal performs more segment browsing, partial rewriting, and frequent switching, while the data archiving host is more inclined to batch reading and sequential verification. Traditional management methods in this scenario are prone to problems such as gradually dispersing logical address distribution, incomplete partial write-back, files related to the same project being scattered into multiple discontinuous segments, increased cross-segment read / write jumps, and long recovery times after abnormal unplugging. In actual use, staff often report that after the same project is completed on one host and then switched to another host, the opening speed of the materials decreases significantly, the cache builds up more slowly when dragging the timeline, disk activity fluctuates intermittently during the export process, and there are even cases where the write-back time is too long after abnormal unloading. These problems directly affect the efficiency of continuous operation.

[0038] To address the aforementioned issues, the method of this invention is deployed in a portable SSD management program in this scenario. When a portable SSD is connected to any host, the system first reads the correspondence between the current logical address and the physical storage block, and simultaneously reads the running state retained before the last unloading. The current logical address segment is then overlaid with completed write boundaries, incomplete write-back positions, address migration termination positions, and organizational segment closure states. Subsequently, continuity detection is performed on adjacent logical address segments in the state overlay queue, marking overlapping boundary positions, address transition positions, and incomplete write-back positions. Then, based on the tail write records, write closure and rearrangement are performed to re-form a regular organizational segment from the continuously available physical storage area. After this processing, the system establishes a virtual storage organization table, providing a unified basis for subsequent access behavior organization. Operationally, no manual intervention is required; after connection, simply opening the project directory normally will allow the background program to automatically complete the organization correction.

[0039] After the virtual storage organization table is formed, the system begins to organize the access records of the currently mounted files. For video master footage, proxy files, subtitle files, and project index files within the same project, the program writes their access actions into the access record queue in the order they occur, and then locates the addresses based on logical address location and organizational segment location. When multiple access records correspond to the same or adjacent organizational segments and appear consecutively within a preset time window, these records are merged into the same access cluster; if access frequently switches between multiple files, it is identified as an intersecting access segment; if the previous write has not been completed and the next write is about to occur, it is identified as a write-back interval segment. In this way, the originally fragmented operation traces are organized into a session access sequence with a sequential relationship. For on-site operators, this means that when frequently used projects are reopened, the system can more easily identify which files need to be loaded first, which files are suitable for being moved forward, and which files are low-activity follow-up objects, thus no longer passively reading all files indiscriminately as in the traditional method.

[0040] After generating the session access sequence, the system inputs it into an improved FEDformer model. The model first divides the session access sequence into multiple continuous subsequences with a fixed step size, then extracts the spectral expansion results using a Fast Fourier Transform (FFT), separating the stably recurring periodic components and locally concentrated burst components. Subsequently, the system sets multiple sliding windows of different lengths for both the periodic and burst component sequences, extracting local segments and calculating the difference between the first and last values ​​of each segment, as well as the change in the slope of the middle segment, to obtain segment change chains at different time scales. These change chains are then arranged uniformly according to their time starting points to form trend trajectory groups. Next, the periodic and burst trajectories are aligned at the same time position to generate a combined trajectory sequence, which is then recursively used to obtain the predicted access sequence. Based on the direction of access intensity change, the direction of access position advancement, and the direction of independent access enhancement in the predicted access sequence, the system determines each file individually, forming queues of files to be moved forward, files to be moved backward, and files to be isolated. On-site, frequently accessed clips, proxy files, and project cache indexes at the editing station are identified as objects to be moved forward, archived materials that are rarely accessed are identified as objects to be moved backward, and proofreading files that are accessed only at specific stations and whose independent access is significantly enhanced are identified as objects to be isolated.

[0041] After obtaining the access migration results, the system performs a migration rearrangement on the virtual storage organization table. For objects to be moved forward, the program scans the preceding segments along the organization segment arrangement direction, eliminating segments with discontinuous addresses and excessive switching records, then writes the qualified free address segments into the forward candidate segment list, and loads them one by one according to the original file arrangement order. For objects to be moved backward, the program searches for a free position further back in the following segments compared to the original segment of the current object, and updates the remaining address range in real time. For objects to be isolated, the program directly allocates the corresponding free address segment from the independent organization segment, separates the file from the original segment, and writes it into the independent segment. After the migration is completed, the system releases the original address space and rewrites the correspondence between the file logical address and the organization segment according to the actual placement relationship, forming a rearranged mapping table. At this point, the actual read and write requests in the current session are scanned sequentially. Discrete read and write requests with small address intervals within the same segment are merged into continuous read and write segments. Cross-segment requests are split into segmented execution queues based on boundaries. Access requests falling into independent segments are transferred to a controlled access queue. Finally, a unified sorting process is used to generate an execution schedule table. In normal use, users will only experience smoother timeline dragging, shorter preview times, and more continuous disk throughput during export, without directly perceiving the complex rearrangement and scheduling performed in the background.

[0042] To verify the effectiveness of this invention, three typical host systems were selected for continuous testing at the production center: a graphics workstation, a portable editing terminal, and a data archiving host. The test hard drive had a capacity of 2TB, with consistent interface standards. The test data consisted of the same batch of project data, including 612GB of main video footage, 188GB of proxy files, 74GB of audio files, 16GB of subtitle and index files, 93GB of stage-exported files, 127GB of cache and temporary files, with the remainder being free space. Two scenarios were set up during the test: one used traditional portable solid-state drive management, employing only a conventional file system and basic cache; the other used the method of this invention for virtual storage organization, session access prediction, migration rearrangement, execution scheduling, and breakpoint recovery. The test ran continuously for 120 hours, covering actual business operations such as continuous reading, continuous writing, alternating calls, abnormal disconnection recovery, and repeated mounting across hosts. To avoid random factors, all core indicators were averaged over multiple rounds.

[0043] Table 1 Comparison of performance in cross-host scenarios ; As shown in Table 1, although the initialization time is slightly longer than the traditional scheme during the first mount due to the need to perform logical address continuity detection, abnormal position correction, and virtual storage organization table construction, the available readiness time is significantly shortened during the second and subsequent mounts. This indicates that the present invention can significantly reduce the cost of repeated reorganization by preserving the running state and rebuilding the organization segments. The first opening time of project files and the second opening time of frequently used materials both show significant decreases, indicating that session access sequence and access migration prediction can accurately identify high-frequency access objects and move them to organization segments more suitable for continuous access. The time to continuous preview when dragging the timeline decreases from 4.7 seconds to 2.8 seconds, indicating that after sequential shaping of read and write requests, the continuous hit rate within the same segment is significantly improved, the number of cross-segment jumps is significantly reduced, and disk access is transformed from discrete to continuous.

[0044] Regarding data security and recovery, traditional management methods typically rely on routine file system checks and variable-length writes after abnormal unmounting, resulting in lengthy recovery times to a readable state and unstable write boundaries. This invention's method continuously encapsulates recovery state units through an execution schedule table. In the event of an abnormal disconnection, it can directly perform tail writes based on the incomplete write-back locations and close the boundaries of unclosed data segments after the completed write boundaries. Therefore, the recovery time to a readable state is reduced to 8.9 seconds, and the incomplete write-back completion time is reduced to 5.4 seconds. After 120 hours of continuous operation, address jumps, boundary overlaps, and the number of unclosed write-back markers all significantly decreased, indicating that this method not only improves speed in the short term but also maintains a relatively stable virtual storage organization state during long-term repeated mounting and frequent cross-host usage.

[0045] This embodiment demonstrates that the present invention addresses the problems of chaotic logical address organization, unusable access behavior, delayed segment migration and adjustment, disordered read / write scheduling, and insufficient recovery continuity of mobile solid-state drives in real-world application environments with multiple hosts, frequent switching, complex access patterns, and a high probability of abnormal interruptions. It establishes a complete processing chain from virtual storage organization table creation, session access sequence construction, access migration prediction, organizational segment migration rearrangement, request order shaping to breakpoint recovery. After actual deployment, this method not only improves the smoothness of common business operations such as opening materials, dragging timelines, continuous exporting, and batch verification, but also significantly improves data recovery efficiency after abnormal disconnections and address organization stability under long-term continuous operation, effectively meeting the management needs of mobile solid-state drives in complex cross-host business scenarios.

[0046] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for managing a portable solid-state drive based on virtual storage, characterized in that, Includes the following steps: Step 1: Read the current logical address arrangement state of the portable solid-state drive and the running state retained before the last unloading, and perform continuity detection and correction to form a virtual storage organization table; Step 2: Based on the virtual storage organization table, locate the address, divide the time window, and extract the segment type of the currently mounted file access records to form a session access sequence; Step 3: Input the session access sequence into the improved FEDformer model, and perform time series prediction through the frequency domain decomposition module, trend coding module, fusion inference module and migration decoding module to obtain the access migration results for the next time period; Step 4: Based on the access migration results, perform migration rearrangement on the virtual storage organization table, and generate a rearrangement mapping table by filtering by segment; Step 5: Based on the rearranged mapping table, perform a sequential scan of the actual read and write requests in the current session to form a continuous read and write segment queue, split the requests to form a segmented execution queue, and generate an execution schedule table; Step 6: Encapsulate the recovery status unit according to the order of issuance recorded in the execution schedule table, and generate the breakpoint recovery table.

2. The method for managing a mobile solid-state drive based on virtual storage according to claim 1, characterized in that, Step one specifically involves: After the external solid-state drive is connected to the host, read the correspondence between the current logical address and the physical storage block, and the running state retained before the last unloading; The starting position, ending position, and order of the physical storage block corresponding to the current logical address segment are superimposed with the running status according to the corresponding logical address segment to form a status superposition queue. Perform continuity detection on adjacent logical address segments in the state overlay queue. The continuity detection is to mark the area where the termination position intersects and overlaps with the starting position of the next logical address segment as the boundary overlap position. The regions where the physical storage block numbers are not continuous between adjacent logical address segments are marked as address transition locations, and the regions where there are unwritten data blocks after the written boundary are marked as unclosed write-back locations, forming an abnormal location marking queue. Based on the abnormal location marker queue, a correction process is performed on each abnormal location. The correction process is to retain the termination boundary of the preceding logical address segment and truncate the overlapping part of the following logical address segment at the boundary overlap location. Extract the empty physical storage block between the address segment before the address jump and the address segment after the address jump position and perform padding and rearrangement. For unclosed write-back positions, call the tail write record retained before the last unloading to perform padding and closure, forming a corrected address queue. The logical address segments in the corrected address queue are reordered according to their physical location. Logical address segments that are physically continuous and do not have overlapping boundaries, address jumps, or unclosed write-backs are merged into the same organization segment. Then, the segment start address, segment end address, and internal arrangement order are written to each organization segment in sequence to form a virtual memory organization table.

3. The method for managing a mobile solid-state drive based on virtual storage according to claim 1, characterized in that, Step two specifically involves: Read the logical address and organization segment of each file currently mounted in the virtual storage organization table, and record each read action, write action and call switch action in the order in which the file access occurs, forming an access record queue; Perform address location processing and time window division on each file access record in the access record queue to obtain an access cluster set; Segment type extraction is performed on each access cluster in the access cluster set. The segment type extraction is to extract consecutive file access records along the organizational segment order within the same access cluster into consecutive access segments. Extract file access records that switch back and forth between two or more files within the same access cluster into cross access segments; extract file access records between the end of the previous write operation and the start of the next write-back operation within the same access cluster into write-back interval segments, forming a set of access segments; According to the distribution order of each access cluster in the virtual memory organization table, the consecutive access segments, intersecting access segments, and write-back interval segments in the access segment set are rearranged. The rearrangement is to place the access segments located in the preceding organization segment before the access segments located in the following organization segment, and then sort the access segments located in the same organization segment according to the actual occurrence time to form the segment sorting result. Based on the segment sorting results, the sorted consecutive access segments, cross access segments, and write-back interval segments are concatenated in chronological order, and segment switching markers and time connection markers are written between adjacent access segments to form a session access sequence.

4. The mobile solid-state drive management method based on virtual storage according to claim 3, characterized in that, The process of performing address location processing and time window division on each file access record in the access record queue to obtain an access cluster set is as follows: Each file access record is mapped to an organization segment in the virtual storage organization table. When multiple file access records correspond to the same or adjacent organization segments, the current multiple file access records are marked as segment adjacency access records. When multiple file access records correspond to different and non-adjacent organizational sections, the current multiple file access records are marked as section jump access records, forming a section location result table; Based on the segment location result table, the time window is divided for each file access record. File access records that appear consecutively within a preset time length are divided into the same time window, and continuous read actions, continuous rewrite actions, and alternating call actions are identified within each time window. When multiple file access records within the same time window simultaneously meet the conditions of segment adjacency and action continuity, the current multiple files are merged into the same access cluster to form an access cluster set.

5. The method for managing a portable solid-state drive based on virtual storage according to claim 1, characterized in that, The improved FEDformer model is specifically as follows: The session access sequence is input into the frequency domain decomposition module, and frequency domain splitting is performed to obtain the decomposed sequence group. The decomposed sequence group is input into the trend encoding module. Multiple sliding windows of different lengths are set for the periodic component sequence and the burst component sequence, and local sequence segments are extracted along the time axis segment by segment using each sliding window. For two adjacent local sequence segments in the same component sequence, perform head-to-tail alignment, calculate the difference in the first value of the segment, the difference in the last value of the segment, and the change in the slope of the middle segment, and then concatenate multiple consecutive local sequence segments according to the window advancement order to obtain the segment change chain at each time scale; Arrange the segments of change at each time scale according to their starting point to form a trend trajectory group; Input the trend trajectory group into the fusion and inference module, align the periodic trajectory and sudden trajectory in the trend trajectory group according to the same time position, and use the periodic trajectory as the main sequence to superimpose the sudden trajectory segment by segment onto the corresponding time position to obtain the combined trajectory sequence. The combined trajectory sequence is recursively pushed forward according to the preset prediction length, and prediction segments at different time positions are generated one by one. The prediction segments are then spliced ​​together in the order of generation to form a prediction access sequence. In the predicted access sequence, the direction of access intensity change, the direction of access position advancement, and the direction of independent access enhancement are scanned segment by segment, and the scan results are merged to form a migration determination group; The migration determination group is input into the migration decoding module. Each determination segment in the migration determination group is expanded according to the corresponding file. Files that are determined to advance the access position and whose access intensity increases beyond the preset threshold are marked for forward movement. Files that have completed the forward movement mark are written into the forward movement queue in sequence. Files that are determined to have been moved backwards in access positions and whose access intensity has decreased by more than a preset threshold are marked as moved backwards, and the files marked as moved backwards are written into the queue to be moved backwards in sequence. Files that are determined to have an increase in independent access counts exceeding a preset threshold and a decrease in cross-segment switching counts exceeding a preset threshold are marked for isolation, and the files that have been marked for isolation are written into the isolation queue in sequence. The queues to be moved forward, to be moved backward, and to be isolated are deduplicated, sorted, and output to form the access migration results.

6. The method for managing a mobile solid-state drive based on virtual storage according to claim 4, characterized in that, The process of inputting the session access sequence into the frequency domain decomposition module and performing frequency domain splitting to obtain the decomposed sequence group is as follows: The session access sequence is divided into multiple continuous subsequences with a fixed step size; For each continuous subsequence, the corresponding spectrum expansion result is obtained by performing a fast Fourier transform. In each spectrum expansion result, each frequency component is scanned in order from low to high frequency position. Frequency components that appear repeatedly in adjacent subsequences are retained as stable components, and frequency components that appear only in local subsequences are extracted as abrupt components. Inverse transformations are performed on the stable components and the mutation components respectively. The stable components after inverse transformation are then spliced ​​together in the original segmentation order to form a periodic component sequence. The mutation components after inverse transformation are then spliced ​​together in the original segmentation order to form a burst component sequence. The periodic component sequences and burst component sequences are aligned according to their time positions to form a decomposed sequence group.

7. The mobile solid-state drive management method based on virtual storage according to claim 1, characterized in that, Step four specifically involves: For each file in the forward set, backward set, and isolated set of access migration results, read the current logical address and the organization segment it belongs to, and record the queue position according to the original arrangement order of each file in the virtual storage organization table to form a migration orchestration queue. Based on the migration orchestration queue, each file in the forward queue is filtered for forward shift segments to form a forward shift placement table; Based on the migration orchestration queue, each file in the move-back queue is filtered for move-back segments to form a move-back placement table; Based on the migration orchestration queue, each file in the isolation queue is stripped and transferred to form an isolation placement table; Each file in the forward placement table, backward placement table, and isolated placement table is read from its original address segment and written to its corresponding forward placement address segment, backward placement address segment, and corresponding independent organization segment address segment. After each file is written, the address position occupied by the original address segment of the current file is released to form the actual placement table. Based on the actual location table, the correspondence between the file logical address and the organizational segment is rewritten according to the actual location relationship after file migration. The new logical address start point, new logical address end point and new organizational segment of each file after migration are written into the mapping record area. The records in the mapping record area are reordered according to the organizational segment arrangement order and the address order of each file in the segment to form a rearranged mapping table.

8. The method for managing a portable solid-state drive based on virtual storage according to claim 1, characterized in that, Step five specifically involves: Read the file logical address range and organizational segment position corresponding to each read / write request in the current session, map each read / write request to the corresponding record in the rearranged mapping table in the order of arrival, and write the request start address, request end address, segment number and request type to each read / write request to form a request mapping queue. The read and write requests in the request mapping queue are scanned sequentially according to the organization segment number. The sequential scan is to concatenate discrete read and write requests with the same segment number and an address interval not exceeding a preset span in sequence, perform segment filling registration on the gap between the end address of the previous read and write request and the start address of the next read and write request, and merge the concatenated request fragments in ascending order of address to form a continuous read and write segment queue. For cross-segment read / write requests in the continuous read / write segment queue, a splitting process is performed. The splitting process involves cutting each cross-segment read / write request into segments according to the boundaries of the covered organizational segments, resulting in multiple sub-request segments corresponding to different organizational segments. Arrange the sub-request segments belonging to the same organizational segment in the original order of request arrival, and write the arranged sub-request segments one by one into the execution sub-queue of the corresponding segment to form a segmented execution queue. Based on the request mapping queue, controlled transfer processing is performed on access requests that fall into an independent organization segment to obtain a controlled access queue; Based on the continuous read / write segment queue, the segmented execution queue, and the controlled access queue, a unified distribution sorting is performed according to the organizational segment arrangement order in the rearranged mapping table to form a distribution order queue. Each consecutive read / write segment, execution subqueue, and controlled access request in the distribution sequence queue is sequentially written into the preset scheduling record area. When writing each item, the corresponding distribution sequence number, the organization segment to which it belongs, the execution start address, the execution end address, and the preceding and following connection positions are written synchronously. All records in the scheduling record area are ordered and arranged according to the distribution sequence number to form an execution scheduling table.

9. A method for managing a portable solid-state drive based on virtual storage according to claim 1, characterized in that, Step six specifically involves: Data reading, data writing, and segmented write-back are executed according to the order of issuance recorded in the execution schedule table. After each round of continuous read / write segment execution or segmented execution queue write-back is completed, the corresponding latest address position, completed write boundary, and incomplete write-back position are recorded and encapsulated into a recovery state unit. When an abnormal disconnection or abnormal unloading is detected, the current writing operation that has not yet been completed is stopped, the incomplete write-back position in the recovery status unit is extracted, the tail writing is performed on the data block corresponding to the incomplete write-back position, the boundary closure is performed on the data segment after the completed write boundary that has not yet been closed, and the writing result and the closure result are written to the recovery retention area to form the breakpoint recovery table. After the external SSD is connected to the host again, the order of data blocks corresponding to the incomplete write-back positions is restored according to the breakpoint recovery table.

Images