SSD (Solid State Disk) cache management method

Abstract

The invention discloses an SSD (Solid State Disk) cache management method, which relates to the technical field of SSDs, and comprises the following steps: step 1, cache space division: the cache space of the SSD is divided into a plurality of different types of sub-cache regions, and each sub-cache region at least comprises a hot data cache region, a cold data cache region and an intermediate data cache region; 2, data popularity monitoring: monitoring the access frequency of the data stored in the SSD in real time, judging the popularity level of the data according to the access frequency, judging the data with the access frequency higher than a set threshold as hot data, judging the data with the access frequency lower than another set threshold as cold data, and judging the data between the hot data and the cold data as intermediate data; 3, data cache allocation: according to the heat level of the data, storing hot data into a hot data cache region, storing cold data into a cold data cache region, and storing intermediate data into an intermediate data cache region; the cache utilization rate and the read-write performance are comprehensively improved, and the delay is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of solid-state drives, and in particular to an SSD cache management method. Background Art

[0002] In the field of computer storage technology, SSDs have become mainstream storage devices due to their high read and write speeds and low power consumption. With the rapid growth of data volumes and users' increasing demands for data processing speeds, the cache management efficiency of SSDs has become increasingly critical to their overall performance.

[0003] From the perspective of cache space division, traditional methods often simply divide the cache space into hot data areas and cold data areas with fixed ratios. This approach lacks flexibility and cannot adaptively adjust based on changes in the actual storage capacity of the SSD solid-state drive and dynamic changes in data access patterns. For example, in some specific application scenarios, such as big data analysis and video editing, the proportion of hot data may increase significantly in a short period of time. However, due to the fixed cache space division, the hot data area cannot be expanded in a timely manner, resulting in a large amount of hot data being unable to be effectively cached, and the cache hit rate frequently decreases, seriously affecting the data read and write speed.

[0004] Traditional methods for monitoring data popularity often fail to accurately reflect real-time data popularity. Many methods rely solely on simple access counts to determine data popularity, ignoring the temporal and spatial locality of data access. For example, data that is periodically accessed may be mistakenly identified as cold data using simple counts. This results in data that should be stored in efficient cache areas being placed in slower-access areas, thereby reducing overall system performance.

[0005] Traditional cache update strategies have significant flaws. Common algorithms, such as the first-in-first-out (FIFO) algorithm, eliminate data based solely on the order in which it enters the cache, completely ignoring the data's actual access frequency and importance. In practice, this can result in important and frequently accessed data being prematurely evicted simply because it entered the cache early. Subsequent accesses to this data require frequent re-reading from primary storage, significantly increasing system I / O overhead and reducing system efficiency.

[0006] Traditional approaches to cache read optimization also fall short. Most traditional approaches lack effective hierarchical search and promotion mechanisms for data of varying popularity. When required data is found in a cold data area, it isn't promoted to a more appropriate cache location. Consequently, subsequent access to the data requires reading from the cold data area, increasing read time and reducing overall read efficiency.

[0007] In summary, existing SSD cache management methods have many drawbacks and are unable to meet the growing data processing needs. There is an urgent need for a more efficient and intelligent cache management method to significantly improve the performance of SSDs. Summary of the Invention

[0008] In order to overcome the above-mentioned shortcomings, the present invention aims to provide a technical solution that can solve the above-mentioned problems.

[0009] A method for managing an SSD cache includes the following steps: Step 1: Cache space division: Divide the cache space of the SSD into multiple sub-cache areas of different types, wherein the sub-cache areas include at least a hot data cache area, a cold data cache area, and an intermediate data cache area; Step 2: Data heat monitoring: This monitors the access frequency of data stored in the SSD in real time and determines the heat level of the data based on the access frequency. Data with an access frequency above a set threshold is considered hot data, data with an access frequency below another set threshold is considered cold data, and data with an access frequency between the two is considered intermediate data. Step 3: Data cache allocation: according to the heat level of the data, hot data is stored in the hot data cache area, cold data is stored in the cold data cache area, and intermediate data is stored in the intermediate data cache area; Step 4: Cache update strategy. When new data is written to the SSD, if the hot data cache is full, the least recently used algorithm is used to migrate the hot data in the hot data cache that has not been accessed for the longest time to the intermediate data cache, and at the same time, the new hot data is written to the hot data cache. If the intermediate data cache is full, the intermediate data in the intermediate data cache that has not been accessed for the longest time is migrated to the cold data cache, and at the same time, the new intermediate data is written to the intermediate data cache. If the cold data cache is full, some cold data in the cold data cache is deleted according to the preset elimination strategy to make room for new data. Step 5: Cache read optimization. When reading data, the hot data cache is searched first. If the data is not found, the intermediate data cache and the cold data cache are searched in that order. If the data is found in the cold data cache, the data is promoted to the intermediate data cache and the data heat level is updated. Step 6: Cache consistency maintenance. When data is updated in the main storage area of the SSD, the corresponding cache data is updated synchronously to ensure the consistency of the cache data with the main storage area data.

[0010] As a further solution of the present invention: in step one, when dividing the cache space, the sizes of the hot data cache area, the cold data cache area and the intermediate data cache area are dynamically adjusted according to the storage capacity of the SSD solid state drive and the historical data access situation.

[0011] As a further solution of the present invention: in step 2, when monitoring the data access frequency, a sliding window algorithm is used to count the number of data accesses in the window at fixed time intervals as the access frequency of the data in the time period.

[0012] As a further solution of the present invention: in the step 4, the preset elimination strategy is based on the storage time of the data, and the cold data with the longest storage time is deleted first.

[0013] As a further solution of the present invention: in step six, a write-back strategy is adopted to maintain cache consistency, that is, when data is modified in the cache, the modification mark is first recorded in the cache, and when the cache data is replaced or the system is idle, the modified data is written to the main storage area.

[0014] As a further solution of the present invention: in step five, when the data is promoted to the intermediate data cache, if the intermediate data cache is full, a least recently used algorithm is first used to eliminate part of the intermediate data, and then the promoted data is written into the intermediate data cache.

[0015] As a further solution of the present invention: it also includes step seven: introducing an artificial intelligence prediction model, using historical data access patterns and real-time monitored access frequency data, and constructing a data access prediction model through a machine learning algorithm to predict data that may be frequently accessed in a certain period of time in the future, and pre-loading data predicted to be hot data from the main storage area to the hot data cache area in advance.

[0016] As a further solution of the present invention: in step seven, the machine learning algorithm includes but is not limited to recurrent neural networks and their variants, long short-term memory networks, and continuously optimizes the accuracy of the prediction model through training on a large amount of historical data.

[0017] As a further solution of the present invention: it also includes step eight: cache defragmentation, regularly defragmenting each sub-cache area, and when the proportion of continuous free storage space in a certain sub-cache area to the total space of the area is lower than a set value, starting the defragmentation program to merge the dispersedly stored data to improve the utilization of the cache space and the data reading and writing efficiency.

[0018] As a further solution of the present invention: in step eight, when defragmenting, a bidirectional linked list data structure is used to record the data storage location, and the linked list pointer is adjusted to achieve rapid merging and sorting of data, thereby reducing the number of data read and write times during the defragmentation process.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1) By dynamically dividing the cache space, the present invention can adaptively adjust the actual storage capacity and data access conditions of the SSD solid-state drive, thereby improving the utilization efficiency of the cache space and effectively avoiding the performance degradation problem caused by unreasonable cache space division. 2) Using a sliding window algorithm for data heat monitoring can more accurately reflect the real-time access pattern of data, making the determination of data heat level more precise, providing a reliable basis for subsequent cache allocation and management, and improving the cache hit rate. 3) The multi-stage cache update strategy and hierarchical search and data promotion mechanism ensure that the most valuable data is always stored in the cache, improving data reading speed and overall system performance. 4) The cache consistency maintenance method of the write-back strategy reduces the number of data writes, reduces write latency, and improves the system's write performance. 5) The introduction of artificial intelligence prediction models can preload data that may be frequently accessed in advance, further improving the cache hit rate, reducing the waiting time for data reading, and improving the user experience. 6) The cache defragmentation mechanism improves cache space utilization and data read and write efficiency by optimizing the data storage structure, thereby extending the service life of the SSD solid state drive.

[0020] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 It is a schematic flow chart of the steps of the present invention. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0024] See also Figure 1In an embodiment of the present invention, a method for managing an SSD cache is provided: 1. Cache space division The SSD cache space is divided into multiple sub-cache areas of different types, including at least a hot data cache area, a cold data cache area, and an intermediate data cache area. During the division, the size of each sub-cache area is dynamically adjusted based on the SSD storage capacity and historical data access. For example, if the number of hot data accesses increases over a period of time, the hot data cache area can be appropriately expanded to improve the cache hit rate of hot data.

[0025] Specifically, first, obtain the SSD's storage capacity and historical data access logs. By analyzing the historical data access logs, calculate the percentages of hot data, cold data, and intermediate data in different time periods. For example, over the past week, the average percentage of hot data was 30%, cold data was 50%, and intermediate data was 20%. Based on these statistics and the current SSD storage capacity, a preliminary ratio of 3:5:2 is determined for the initial sizes of the hot data cache, cold data cache, and intermediate data cache. During system operation, data access is continuously monitored. If, within a certain period, the number of hot data accesses suddenly increases, causing the hit rate of the hot data cache to decrease, the system will reassess cache space allocation according to pre-set rules. For example, if the actual percentage of hot data has reached 40%, the system will appropriately reduce the size of the cold data cache and intermediate data cache, adjusting the ratio of the hot data cache to 4:4:2 to accommodate the dynamic changes in data access.

[0026] 2. Data heat monitoring Real-time monitoring of data access frequency stored on SSDs. Using a sliding window algorithm, the number of data accesses within a window is counted at fixed intervals as the data access frequency within that time period. Data heat level is determined based on access frequency: data with an access frequency above a set threshold is considered hot data, data with an access frequency below another set threshold is considered cold data, and data with an access frequency in between is considered intermediate data. This approach more accurately reflects the real-time heat level of data, providing a reliable basis for subsequent cache allocation and management.

[0027] Specifically, a sliding window is set up with a fixed interval of 10 seconds. Within each window, the system records the number of accesses to all data. For example, in the first 10-second window, file A was accessed 5 times, file B was accessed 2 times, and file C was accessed 0 times. By counting the number of accesses within the window and combining it with pre-set hot and cold data thresholds (assuming the hot data threshold is 4 times / 10 seconds and the cold data threshold is 1 time / 10 seconds), file A is determined to be hot data, file B to be intermediate data, and file C to be cold data. As time passes, the sliding window continues to move, and data access statistics within new 10-second windows are included in the statistics. For example, in the next 10-second window, file A is accessed 3 times, file B to be accessed 4 times, and file C to be accessed 1 time. At this point, the data heat level is reassessed, and file A becomes intermediate data, file B becomes hot data, and file C remains cold data. This sliding window algorithm accurately monitors changes in data heat level in real time.

[0028] 3. Data cache allocation Based on the data's heat level, hot data is stored in the hot data cache, cold data in the cold data cache, and intermediate data in the intermediate data cache. This allocation ensures that frequently accessed data is stored in the hot data cache, which has the fastest read and write speeds, improving data read efficiency. Meanwhile, less frequently accessed cold data is stored in the slower but larger cold data cache, effectively utilizing cache space.

[0029] Specifically, when the system determines that certain data is hot data, such as file B in the above example is determined to be hot data at a certain moment, the system stores it in the hot data cache. The hot data cache uses high-speed storage media to ensure fast reading and writing of hot data. For files determined to be cold data, such as file C, the system stores them in the cold data cache. The storage capacity of the cold data cache is relatively large, but the reading and writing speeds are relatively slow. For intermediate data, such as file A, which is intermediate data at a certain stage, it is stored in the intermediate data cache. The reading and writing speeds and storage capacity of the intermediate data cache are between the hot data cache and the cold data cache. Through this precise cache allocation based on the data heat level, the overall performance of the cache is improved.

[0030] 4. Cache Update Strategy When new data is written to the SSD, if the hot data cache is full, the least recently used (LRU) algorithm is used to migrate the hot data in the hot data cache that has not been accessed for the longest time to the intermediate data cache, and the new hot data is written to the hot data cache. If the intermediate data cache is full, the intermediate data in the intermediate data cache that has not been accessed for the longest time is migrated to the cold data cache, and the new intermediate data is written to the intermediate data cache. If the cold data cache is full, some cold data in the cold data cache is deleted according to the preset elimination strategy (based on the data storage age, the cold data with the longest storage age is deleted first) to make room for new data. This multi-stage cache update strategy can effectively manage cache space and ensure that the most valuable data is always stored in the cache.

[0031] Specifically, assuming the hot data cache has a capacity of 100MB, when new hot data is to be written and 95MB of hot data cache space is already occupied, the system initiates a cache update strategy. Using a least-repeated (LRU) algorithm, the system searches for the hot data cache's least recently accessed data. For example, if file D has not been accessed in the hot data cache for one minute, while other hot data has been accessed to varying degrees within that minute, the system migrates file D to the intermediate data cache and then writes the new hot data into the hot data cache. If the intermediate data cache is full, for example, if the intermediate data cache has a capacity of 80MB and is fully occupied, the system also uses the least-repeated (LRU) algorithm to migrate the least recently accessed intermediate data in the intermediate data cache (e.g., file E) to the cold data cache and then writes the new intermediate data into the intermediate data cache. If the cold data cache is full, the system uses a data age-based eviction strategy to delete the oldest cold data (e.g., file F, which has been in the cold data cache for one hour, while all other cold data have been stored for less than one hour) to make room for the new data.

[0032] 5. Cache Read Optimization When reading data, the hot data cache is searched first. If the data is not found, the intermediate data cache and then the cold data cache are searched. If the data is found in the cold data cache, it is promoted to the intermediate data cache and the data's heat level is updated. Furthermore, when promoting data to the intermediate data cache, if the intermediate data cache is full, the LRU algorithm is used to eliminate some of the intermediate data before the promoted data is written to the intermediate data cache. This hierarchical search and data promotion mechanism can improve data reading speed and reduce the time overhead of data reading.

[0033] Specifically, when the system receives a request to read file G, it first searches in the hot data cache. If not found, it searches in the intermediate data cache. Assuming that file G is found in the intermediate data cache, the system reads file G and returns it to the user. At the same time, the system promotes file G to the hot data cache and updates its heat level. If file G is not found in the intermediate data cache, it continues to search in the cold data cache. If file G is found in the cold data cache, the system reads it and returns it to the user, then promotes file G to the intermediate data cache and updates its heat level. When promoting file G to the intermediate data cache, if the intermediate data cache is full, the system first uses the LRU algorithm to eliminate some intermediate data (such as file H, which is the file in the intermediate data cache that has not been accessed for the longest time), and then writes file G to the intermediate data cache. Through this hierarchical search and data promotion mechanism, the data reading efficiency is improved.

[0034] 6. Cache consistency maintenance When data is updated in the main storage area of the SSD, the corresponding cache data is updated synchronously to ensure consistency between the cached data and the main storage area. A write-back strategy is used to maintain cache consistency. When data is modified in the cache, a modification marker is first recorded in the cache. The modified data is then written to the main storage area when the cached data is replaced or the system is idle. This strategy reduces the number of data writes, lowers write latency, and improves overall system performance.

[0035] Specifically, when file I in the primary storage area is updated, the system immediately searches for the file in the corresponding cache area (assuming file I was originally stored in the hot data cache). Upon finding the file, a modification marker is first recorded in the cache. At this point, file I in the cache is in a modified but not yet written to the primary storage area. When the hot data cache area performs a data replacement operation (such as when file J is eliminated using the LRU algorithm) or the system is idle (such as when CPU utilization is below 10% for 5 seconds), the system writes the modified file I in the cache to the primary storage area, completing cache consistency maintenance. This write-back strategy reduces frequent data write operations and improves system performance.

[0036] 7. Cache Defragmentation Each sub-cache area is regularly defragmented. When the proportion of continuous free storage space in a sub-cache area to the total space in that area falls below a set value, the defragmentation process is initiated. During the defragmentation process, a doubly linked list data structure is used to record data storage locations. By adjusting the linked list pointers, data can be quickly merged and organized, reducing the number of data reads and writes during the defragmentation process. Cache defragmentation improves cache space utilization and data read and write efficiency.

[0037] Specifically, the system checks each sub-cache area every hour. Assuming the total capacity of the hot data cache is 100MB, defragmentation is initiated when the proportion of contiguous free storage space in the hot data cache falls below 20% (i.e., less than 20MB of free space). In the hot data cache, a doubly linked list data structure is used to record the storage location of each data block. Adjacent data blocks are found by traversing the doubly linked list, and these adjacent blocks are merged into a single contiguous storage block by adjusting the linked list pointers. For example, if data blocks 1, 2, and 3 were originally stored separately, they can be merged into a single, contiguous, large storage block by adjusting the linked list pointers. During the merging process, data read and write operations are minimized, with data moves performed only when necessary. This defragmentation method improves space utilization in the hot data cache, thereby enhancing data read and write efficiency. The same method is applied to defragmentation in the cold data cache and intermediate data cache.

[0038] The SSD solid-state hard drive cache management method proposed in the present invention has built a comprehensive and efficient cache management system through innovative dynamic partitioning of cache space, precise data heat monitoring, intelligent cache update and read optimization strategy, efficient cache consistency maintenance mechanism, introduction of advanced artificial intelligence prediction model and reasonable cache defragmentation means. This method has innovated the traditional SSD solid-state hard drive cache management method from multiple dimensions, effectively overcomes many drawbacks in the existing technology, and significantly improved the performance of SSD solid-state hard drives in data reading and writing performance, cache space utilization and data processing efficiency. In the current environment where the amount of data continues to grow at a high speed and users have increasingly stringent requirements on storage device performance, the cache management method of the present invention has broad application prospects and huge practical value, and is expected to promote SSD solid-state hard drive cache management technology to a new stage of development and provide strong support for the further development of the computer storage field.

[0039] 8. Introducing AI prediction models Leveraging historical data access patterns and real-time access frequency data, a data access prediction model is constructed using machine learning algorithms (including but not limited to recurrent neural networks and their variant, long short-term memory networks). This model predicts data likely to be frequently accessed in the future and preloads predicted hot data from the primary storage area into the hot data cache. This predictive preloading mechanism further improves cache hit rates and reduces data read latency.

[0040] Specifically, the system collects historical data access logs from the past month, including information such as data access time, access frequency, and file size. Leveraging this data, a data access prediction model is constructed using a recurrent neural network (RNN) and its variant, the long short-term memory network (LSTM). During model training, model parameters such as the learning rate and the number of hidden layer nodes are continuously adjusted to improve model accuracy. After training on a large amount of data, the model can predict data that is likely to be frequently accessed in the future based on real-time access frequency data and historical data access patterns. For example, the model predicts that file K is likely to be frequently accessed within the next 30 minutes. The system preloads file K from the main storage area into the hot data cache. Subsequent accesses to file K can be quickly read directly from the hot data cache, significantly improving cache hit rates and data read speeds.

[0041] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations that come within the meaning and range of equivalents of the claims be embraced therein.

Claims

1. A SSD solid state drive cache management method, characterized in that: The following steps are involved: Step 1: Cache space division: Divide the cache space of the SSD into multiple sub-cache areas of different types, wherein the sub-cache areas include at least a hot data cache area, a cold data cache area, and an intermediate data cache area; Step 2: Data heat monitoring: This monitors the access frequency of data stored in the SSD in real time and determines the heat level of the data based on the access frequency. Data with an access frequency above a set threshold is considered hot data, data with an access frequency below another set threshold is considered cold data, and data with an access frequency between the two is considered intermediate data. Step 3: Data cache allocation: according to the heat level of the data, hot data is stored in the hot data cache area, cold data is stored in the cold data cache area, and intermediate data is stored in the intermediate data cache area; Step 4: Cache update strategy. When new data is written to the SSD, if the hot data cache is full, the least recently used algorithm is used to migrate the hot data in the hot data cache that has not been accessed for the longest time to the intermediate data cache, and at the same time, the new hot data is written to the hot data cache. If the intermediate data cache is full, the intermediate data in the intermediate data cache that has not been accessed for the longest time is migrated to the cold data cache, and at the same time, the new intermediate data is written to the intermediate data cache. If the cold data cache is full, some cold data in the cold data cache is deleted according to the preset elimination strategy to make room for new data. Step 5: Cache read optimization. When reading data, the hot data cache is searched first. If the data is not found, the intermediate data cache and the cold data cache are searched in that order. If the data is found in the cold data cache, the data is promoted to the intermediate data cache and the data heat level is updated. Step 6: Cache consistency maintenance. When data is updated in the main storage area of the SSD, the corresponding cache data is updated synchronously to ensure the consistency of the cache data with the main storage area data.

2. The SSD cache management method according to claim 1, wherein: In step 1, when dividing the cache space, the sizes of the hot data cache area, the cold data cache area, and the intermediate data cache area are dynamically adjusted according to the storage capacity of the SSD solid state drive and historical data access conditions.

3. The SSD cache management method according to claim 1, wherein: In the second step, when monitoring the data access frequency, a sliding window algorithm is used to count the number of data accesses within the window at fixed time intervals as the data access frequency within the time period.

4. The SSD cache management method according to claim 1, wherein: In the step 4, the preset elimination strategy is based on the storage time of the data, and the cold data with the longest storage time is deleted first.

5. The SSD cache management method according to claim 1, wherein: In step six, a write-back strategy is used to maintain cache consistency, that is, when data is modified in the cache, a modification mark is first recorded in the cache, and when the cached data is replaced or the system is idle, the modified data is written to the main storage area.

6. The SSD cache management method according to claim 1, wherein: In step five, when promoting data to the intermediate data cache, if the intermediate data cache is full, a least recently used algorithm is first used to eliminate part of the intermediate data, and then the promoted data is written into the intermediate data cache.

7. The SSD cache management method according to claim 1, wherein: It also includes step seven: cache defragmentation, which regularly defragments each sub-cache area. When the proportion of continuous free storage space in a sub-cache area to the total space of the area is lower than the set value, the defragmentation program is started to merge the scattered stored data to improve the utilization of the cache space and data reading and writing efficiency.

8. The SSD cache management method according to claim 7, characterized in that: In step seven, when performing defragmentation, a bidirectional linked list data structure is used to record the data storage location, and the linked list pointer is adjusted to achieve rapid merging and sorting of data, thereby reducing the number of data read and write times during the defragmentation process.

9. The SSD cache management method according to claim 7, wherein: It also includes step eight: introducing an artificial intelligence prediction model, using historical data access patterns and real-time monitored access frequency data, and building a data access prediction model through a machine learning algorithm to predict data that may be frequently accessed in the future, and pre-load the data predicted to be hot data from the main storage area to the hot data cache area in advance.

10. The SSD cache management method according to claim 9, characterized in that: In step eight, the machine learning algorithm includes but is not limited to a recurrent neural network and its variant, a long short-term memory network, which continuously optimizes the accuracy of the prediction model by training on a large amount of historical data.

Images