WO2021213278A1

WO2021213278A1 - File prefetching method, storage device, and prefetching apparatus

Info

Publication number: WO2021213278A1
Application number: PCT/CN2021/087840
Authority: WO
Inventors: 江舟; 向贵东; 刘金虎
Original assignee: 华为技术有限公司
Priority date: 2020-04-20
Filing date: 2021-04-16
Publication date: 2021-10-28
Also published as: CN113535658A

Abstract

A file prefetching method, a storage device, and a prefetching apparatus. In the file prefetching method, a storage device generates a first keyword according to file access information carried in a read request (101); generates a target keyword using the first keyword and a target access rule template (102); and prefetches a prediction file indicated by the target keyword into a cache (103). The target keyword and the first keyword match the same characteristics. The first keyword and the target keyword match the same characteristics, and therefore, the prediction file indicated by the target keyword may be a file that a user needs to read at the next moment. Therefore, the file prefetching method can be used to predict a file that a processor accesses for the first time, so as to implement the prefetching of a file that has not been accessed, thereby facilitating the processor to increase the hit rate of a read request when accessing a new file.

Description

File prefetching method, storage equipment and prefetching device

Technical field

The embodiments of the present application relate to the field of computer storage, and in particular, to a file prefetching method, storage device, and prefetching device.

Background technique

Cached data prefetching technology is a computer operating system cache storage optimization technology. The cache data prefetching technology refers to loading the aforementioned data from the main memory to the cache memory in advance before the processor in the storage device accesses the data for calculation, so as to increase the hit rate of the read request issued by the host and reduce the processor access to the data. Pause time, so as to achieve the purpose of improving read performance.

In the prior art, the storage device will generate the access rule of the accessed file based on the historical access record to form a historical access queue composed of multiple files. When a certain file in the aforementioned historical access queue is accessed again, the storage device can prefetch the file located after the file in the historical access queue into the cache to increase the hit rate of read requests issued by the host in the storage device , Improve system read performance.

In such a solution, since the aforementioned historical access queue is generated based on files that have been accessed, the storage device can only prefetch files that have been accessed, but cannot prefetch files that have not been accessed. Therefore, it is not conducive for the host to increase the hit rate of read requests when accessing new files in the storage device.

Summary of the invention

The embodiments of the present application provide a file prefetching method and storage device, which are used to predict the file that the host accesses for the first time, so as to realize the prefetching of the files that have not been accessed, which is beneficial to the host when accessing new files. The hit rate of read requests.

In the first aspect, an embodiment of the present application provides a file prefetching method, which is applied to a storage device and is used to prefetch a file from a low-speed storage medium to a high-speed storage medium. In the file prefetching method, the storage device first receives a read request issued by the host, the read request is used to indicate the file to be accessed, and the read request carries file access information of the file. The storage device generates the first keyword according to the file access information. Then, the storage device uses the first keyword and the target access rule template to generate the target keyword. Wherein, the target keyword and the first keyword match the same characteristics, and the target keyword is used to indicate the prediction file. Then, the storage device prefetches the prediction file indicated by the target keyword into the cache.

Wherein, the aforementioned file access information is information related to the file indicated by the read request, for example, file name, file type, access time, file creator name, file visitor name, and access directory. The aforementioned low-speed storage medium may be a cache, and the aforementioned high-speed storage medium may be a hard disk.

In the embodiment of the present application, the target keyword can be generated by using the first keyword and the target access rule template, and then the prediction file to be prefetched is determined based on the target keyword. Since the first keyword and the target keyword match the same characteristics, it can be considered that the predicted file indicated by the target keyword is the file that the user needs to read at the next moment. Therefore, even if it is determined that the file access information of the first keyword comes from the file accessed for the first time, the storage device can use the aforementioned first keyword and the target access rule template to calculate the predicted file that has not been accessed, and calculate the predicted file Prefetch into the cache. It is helpful to improve the hit rate of the host acquiring data in the cache in the storage device.

According to the first aspect, in the first implementation manner of the first aspect of the embodiments of the present application, the storage device stores a plurality of access rule templates, and the target access rule template is the first keyword among the plurality of access rule templates. Corresponding access rule templates, each access rule template is obtained by training multiple initial keywords using a training model based on text semantics.

It can also be understood that the storage device uses a training model based on text semantics to train multiple initial keywords to obtain the aforementioned access rule template. Similarly, the storage device can train multiple access rule templates. Wherein, the access rule template corresponding to the aforementioned first keyword is the aforementioned target access rule template.

In this embodiment, it is proposed that in order to be able to determine the predicted file more accurately, the storage device needs to select the access rule template corresponding to the first keyword (that is, the target access rule template), so that the target key for finding the predicted file is generated. word. In such an embodiment, selecting the corresponding access rule template according to the first keyword is beneficial to improve the accuracy of the predicted file, and further helps to increase the probability that the predicted file is hit by the issued read request. Prefetching the predicted file into the cache is beneficial to improve the hit rate of the host acquiring data in the cache of the storage device.

According to the first implementation manner of the first aspect, in the second implementation manner of the first aspect of the embodiments of the present application, the target access rule template includes the first keyword and/or the first characteristic related word, and the first characteristic related word It is used to indicate the characteristics of the first keyword.

In this embodiment, what is a target access rule template is further proposed, which provides a basis for the correspondence between the first keyword and the target access rule template.

According to the second implementation manner of the first aspect, in the third implementation manner of the first aspect of the embodiments of the present application, the target access rule template further includes a mapping relationship, and the mapping relationship is used to indicate the relationship between the plurality of initial keywords The multiple initial keywords include the first keyword. Using the first keyword and the target access rule template to generate the target keyword includes: taking the first keyword as the input of the mapping relationship, calculating the first keyword using the mapping relationship, and outputting the target keyword , The target keyword meets the characteristics of the first keyword indicated by the first characteristic associated word.

In this embodiment, it is proposed that the target access rule template further includes a mapping relationship, which is generated when the foregoing multiple initial keywords are trained. The mapping relationship is used to indicate the association mode between the multiple initial keywords. It can be understood that when one of the multiple initial keywords and the mapping relationship are known, the storage device can either calculate The remaining initial keywords. Therefore, when the first keyword corresponds to the target access rule template, the first keyword is also applicable to the mapping relationship in the target access rule template. At this time, the storage device may use the first keyword as the input of the mapping relationship, and use the mapping relationship to calculate the first keyword, thereby outputting the target keyword. In such an implementation manner, the specific method for determining the target keyword is clarified, and the reliability of the solution is improved.

According to the second implementation manner of the first aspect or the third implementation manner of the first aspect, in the fourth implementation manner of the first aspect of the embodiments of the present application, when the storage device uses the first keyword to access from multiple When searching for the target access rule template in the rule template, the storage device will perform the following steps:

Determine whether the access pattern template contains the first keyword;

If the access rule template contains the first keyword, it is determined that the access rule template is the target access rule template;

If the access pattern template does not include the first keyword, it is further determined whether the first keyword meets the characteristics indicated by the characteristic associated words in the access pattern template;

If the first keyword conforms to the characteristics indicated by the characteristic associated words in the visit rule template, the visit rule template is determined to be the target visit rule template.

According to the second implementation manner of the first aspect or the third implementation manner of the first aspect, in the fifth implementation manner of the first aspect of the embodiments of the present application, the file access information carried in the read request is also used to generate a second Keywords, both the second keyword and the first keyword are keywords that match the target access rule template. When the storage device simultaneously uses the first keyword and the second keyword to find a target access pattern template from multiple access pattern templates, the storage device will perform the following steps:

Determine whether the access pattern template contains the first keyword and the second keyword;

If the access pattern template includes the first keyword and the second keyword, determine that the access pattern template is the target access pattern template;

If the access pattern template only contains one of the first keyword and the second keyword or does not contain the first keyword and the second keyword, then it is further determined whether the first keyword and the second keyword are All conform to the characteristics indicated by the characteristic associated words in the access rule template;

If both the first keyword and the second keyword conform to the characteristics indicated by the characteristic associated words in the access rule template, the access rule template is determined to be the target access rule template.

According to the first aspect, the first implementation manner of the first aspect to the fifth implementation manner of the first aspect, in the sixth implementation manner of the first aspect of the embodiments of the present application, the file access information is from different access directories File information, the multiple predicted files are located in different access directories.

In this embodiment, it is proposed that since the file access information of the first keyword is determined to be information from files in different access directories, the predicted files predicted by the aforementioned method based on the first keyword may also be located in different access directories. In such an embodiment, compared to only predicting files in the same access directory, the storage range of the prediction file is expanded, so that the prefetched prediction file can come from different access directories.

According to the first aspect and the first implementation manner of the first aspect to the fifth implementation manner of the first aspect, in the seventh implementation manner of the first aspect of the embodiments of the present application, the storage device may also perform access rule template collections. After updating, the access rule template set refers to the aforementioned multiple access rule templates located in the storage device.

In this embodiment, the access rule template set includes one target access rule template and at least one candidate access rule template. Specifically, the storage device determines the degree of association between the target access rule template and each candidate access rule template in at least one candidate access rule template, where the association degree is used to indicate the mapping relationship between the target access rule template and the candidate access rule template. The degree of similarity between the mapping relationships in the candidate access pattern template. Then, the storage device merges the candidate visit rule template corresponding to the degree of association higher than the preset value with the target candidate visit rule template to obtain an updated set of visit rule templates.

In this embodiment, it is proposed that the access rule template set in the storage device can be updated based on the target access rule template corresponding to the first keyword, that is, the storage device updates the predicted file based on the first keyword. A collection of access rule templates in the storage device. Such an implementation manner can enable each access rule template in the set of multiple updated access rule templates to more accurately determine a more accurate prediction file based on the input keywords, so that the aforementioned prediction file can be prefetched to the cache. Improve the hit rate of read requests issued by the host.

In a second aspect, an embodiment of the present application provides a storage device, which includes: a cache, a hard disk, and at least one processor. Among them, at least one processor is configured to perform the following operations: generate a first keyword according to the file access information carried in the read request; use the first keyword and the target access rule template to generate a target keyword, the target keyword and the first keyword A keyword meets the same characteristics, and the target keyword is used to indicate a prediction file; the prediction file indicated by the target keyword is prefetched from the hard disk to the cache.

The file access information is information related to the file indicated by the read request, for example, file name, file type, access time, file creator name, file visitor name, and access directory.

In the embodiment of the present application, the target keyword can be generated by using the first keyword and the target access rule template, and then the prediction file to be prefetched is determined based on the target keyword. Since the first keyword and the target keyword match the same characteristics, it can be considered that the predicted file indicated by the target keyword is a file that the user needs to read at the next moment. Therefore, even if it is determined that the file access information of the first keyword comes from the file accessed for the first time, the storage device can use the aforementioned first keyword and the target access rule template to calculate the predicted file that has not been accessed, and calculate the predicted file Prefetch into the cache. It is helpful to improve the hit rate of the host acquiring data in the cache in the storage device.

According to the second aspect, in the first implementation manner of the second aspect of the embodiments of the present application, the storage device stores a plurality of access rule templates, and the target access rule template is the first keyword among the plurality of access rule templates. Corresponding access rule templates, each access rule template is obtained by training multiple initial keywords using a training model based on text semantics.

According to the first implementation manner of the second aspect, in the second implementation manner of the second aspect of the embodiments of the present application, the target access rule template includes the first keyword and/or the first characteristic related word, and the first characteristic related word It is used to indicate the characteristics of the first keyword.

According to the second implementation manner of the second aspect, in the third implementation manner of the second aspect of the embodiments of the present application, the target access rule template further includes a mapping relationship, and the mapping relationship is used to indicate the relationship between the plurality of initial keywords The multiple initial keywords include the first keyword. The at least one processor is specifically configured to use the first keyword as the input of the mapping relationship, calculate the first keyword using the mapping relationship, and output the target keyword, the target keyword conforming to the first feature The characteristics of the first keyword indicated by the associated word.

According to the second implementation manner of the second aspect or the third implementation manner of the second aspect, in the fourth implementation manner of the second aspect of the embodiments of the present application, the at least one processor is further configured to: determine the access rule template Whether the first keyword is included; if the access rule template contains the first keyword, the access rule template is determined to be the target access rule template; if the access rule template does not include the first keyword, then the access rule template is further judged Whether the first keyword meets the characteristics indicated by the characteristic associated words in the visit rule template; if the first keyword meets the characteristics indicated by the characteristic associated words in the visit rule template, then the visit rule template is determined to be the target visit rule template.

According to the second implementation manner of the second aspect or the third implementation manner of the second aspect, in the fifth implementation manner of the second aspect of the embodiments of the present application, the file access information carried in the read request is also used to generate a second Keywords, the second keyword and the first keyword are keywords that match the target access pattern template; the at least one processor is also used to: determine whether the access pattern template contains the first keyword and the The second keyword; if the access pattern template contains the first keyword and the second keyword, the access pattern template is determined to be the target access pattern template; if the access pattern template only contains the first keyword and the One of the second keywords or does not include the first keyword and the second keyword, then it is further determined whether the first keyword and the second keyword are both in line with the characteristics indicated by the characteristic related words in the access rule template ; If both the first keyword and the second keyword meet the characteristics indicated by the feature associated words in the access rule template, then the access rule template is determined to be the target access rule template.

According to the second aspect and the first implementation manner of the second aspect to the fifth implementation manner of the second aspect, in the sixth implementation manner of the second aspect of the embodiments of the present application, the file access information is from different access directories File information, the multiple predicted files are located in different access directories.

According to the second aspect and the first implementation manner of the second aspect to the fifth implementation manner of the second aspect, in the seventh implementation manner of the second aspect of the embodiments of the present application, the at least one processor further includes a first processing The first processor is configured to use the text semantic-based training model to train the multiple initial keywords to obtain the access rule template.

According to the seventh implementation manner of the second aspect, in the eighth implementation manner of the second aspect of the embodiments of the present application, the first processor may also update the access rule template set, and the access rule template set refers to the aforementioned set of templates located in the Multiple access regularity templates in the storage device.

In this embodiment, the visit rule template set includes one target visit rule template and at least one candidate visit rule template. Specifically, the first processor is configured to determine the degree of association between the target visit rule template and each of the at least one candidate visit rule template, wherein the degree of association is used to indicate the target visit rule template The degree of similarity between the mapping relationship in and the mapping relationship in the candidate access rule template. The first processor is further configured to merge the candidate visit rule template corresponding to the degree of association higher than the preset value with the target candidate visit rule template to obtain an updated set of visit rule templates.

In this embodiment, it is proposed that the access rule template set in the storage device can be updated based on the target access rule template corresponding to the first keyword, that is, at least one processor in the storage device determines that the access rule template is based on the first keyword. After predicting the file, the first processor in the storage device then updates the access rule template set in the storage device. Such an implementation manner can enable each access rule template in the set of multiple updated access rule templates to more accurately determine a more accurate prediction file based on the input keywords, so that the aforementioned prediction file can be prefetched to the cache. Improve the hit rate of the host acquiring data in the cache in the storage device.

In a third aspect, an embodiment of the present application provides a prefetching device, the prefetching device is located in a storage device, and the storage device further includes a cache and a hard disk. The storage device stores a computer program or instruction, and the prefetching device invokes the computer program or instruction to execute the following modules: a keyword generation module for generating the first keyword according to the file access information carried in the read request; a calculation module for Use the first keyword and the target access rule template to generate a target keyword, where the target keyword and the first keyword match the same characteristics, and the target keyword is used to indicate the prediction file; the data migration module is used to transfer The predicted file indicated by the target keyword is prefetched from the hard disk to the cache.

According to the third aspect, in the first implementation manner of the third aspect of the embodiments of the present application, the prefetching device stores a plurality of access rule templates, and the target access rule template is the first key among the plurality of access rule templates. The access rule template corresponding to the word, each access rule template is obtained by training multiple initial keywords using a training model based on text semantics.

According to the first implementation manner of the third aspect, in the second implementation manner of the third aspect of the embodiments of the present application, the target access rule template includes the first keyword and/or the first characteristic related word, and the first characteristic related word It is used to indicate the characteristics of the first keyword.

According to the second implementation manner of the third aspect, in the third implementation manner of the third aspect of the embodiments of the present application, the target access rule template further includes a mapping relationship, and the mapping relationship is used to indicate the relationship between the plurality of initial keywords The multiple initial keywords include the first keyword; the calculation module is specifically configured to use the first keyword as the input of the mapping relationship, and use the mapping relationship to calculate the first keyword, The target keyword is output, and the target keyword meets the characteristics of the first keyword indicated by the first characteristic associated word.

According to the second implementation manner of the third aspect or the third implementation manner of the third aspect, in the fourth implementation manner of the third aspect of the embodiments of the present application, the calculation module is further used to determine whether the access rule template contains The first keyword; if the access pattern template contains the first keyword, the access pattern template is determined to be the target access pattern template; if the access pattern template does not contain the first keyword, the first keyword is further determined Whether it conforms to the characteristics indicated by the characteristic related words in the visit rule template; if the first keyword matches the characteristics indicated by the characteristic related words in the visit rule template, then the visit rule template is determined to be the target visit rule template.

According to the second implementation manner of the third aspect or the third implementation manner of the third aspect, in the fifth implementation manner of the third aspect of the embodiments of the present application, the file access information carried in the read request is also used to generate a second Keywords, the second keyword and the first keyword are both keywords that match the target access rule template; the calculation module is also used to determine whether the access rule template contains the first keyword and the second keyword Keywords; if the visit rule template contains the first keyword and the second keyword, then the visit rule template is determined to be the target visit rule template; if the visit rule template only contains the first keyword and the second keyword If one of the keywords or does not include the first keyword and the second keyword, it is further determined whether the first keyword and the second keyword both conform to the characteristics indicated by the characteristic associated words in the access rule template; if Both the first keyword and the second keyword conform to the characteristics indicated by the characteristic associated words in the visit rule template, and the visit rule template is determined to be the target visit rule template.

According to the third aspect and the first implementation manner of the third aspect to the fifth implementation manner of the third aspect, in the sixth implementation manner of the third aspect of the embodiments of the present application, the file access information is from different access directories File information, the multiple predicted files are located in different access directories.

According to the third aspect and the first implementation manner of the third aspect to the fifth implementation manner of the third aspect, in the seventh implementation manner of the third aspect of the embodiments of the present application, the prefetching device further includes: a template generation module , Is used to train the multiple initial keywords using the text semantic training model to obtain the access rule template.

According to the seventh implementation manner of the third aspect, in the eighth implementation manner of the third aspect of the embodiments of the present application, the template generation module is also used to update the access rule template set. The access rule template set refers to the aforementioned location Multiple access rule templates in the storage device. The visit rule template set includes one target visit rule template and at least one candidate visit rule template.

The template generation module is specifically configured to determine the degree of association between the target access rule template and each of the at least one candidate access rule template, and compare the candidate access rule template corresponding to an association degree higher than a preset value Merge with the target candidate visit rule template to obtain an updated set of visit rule templates.

Wherein, the degree of association is used to indicate the degree of similarity between the mapping relationship in the target visit rule template and the mapping relationship in the candidate visit rule template.

In a fourth aspect, an embodiment of the present application provides a smart chip, the smart chip is located in the storage device in the foregoing embodiment, and the smart chip is used to train the input sample data to output a prediction model. For example, the smart chip is located in the first processor in the seventh implementation of the third aspect of the embodiments of the present application, and the smart chip may use the text semantic-based training model to train the multiple initial keywords to obtain access Regular template. For another example, the smart chip is located in the template generation module in the seventh implementation of the fourth aspect of the embodiments of the present application, and the smart chip may use the text-based semantic training model to train the multiple initial keywords to obtain access Regular template. The smart chip can also update the access rule template collection.

It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

Description of the drawings

In order to more clearly describe the technical solutions of the embodiments of the present application, the accompanying drawings required in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application.

FIG. 1A is a system architecture diagram to which the file prefetching method in an embodiment of the application is applicable;

FIG. 1B is a flowchart of a file prefetching method in an embodiment of this application;

2A is a schematic diagram of an embodiment of file access information in an embodiment of this application;

2B is a schematic diagram of another embodiment of file access information in an embodiment of this application;

FIG. 3 is another flowchart of the file prefetching method in the embodiment of the application;

FIG. 4 is a schematic diagram of an embodiment of a storage device in an embodiment of the application;

FIG. 5 is a schematic diagram of an embodiment of a prefetching device in an embodiment of the application.

Detailed ways

The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments.

The terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of this application and the above-mentioned drawings are used to distinguish similar objects, without having to use To describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments described herein can be implemented in a sequence other than the content illustrated or described herein. In addition, the terms "including" and "having" and any variations of them are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those clearly listed. Those steps or units may include other steps or units that are not clearly listed or are inherent to these processes, methods, products, or equipment.

The following first introduces the applicable system architecture and application scenarios of the file prefetching method proposed in the embodiments of the present application:

As shown in FIG. 1A, it is a system architecture diagram provided by an embodiment of this application. The storage system provided in this embodiment includes a host 01, a controller 00, and multiple hard disks 02. Among them, the host 01 and the controller 00 communicate through a network file system (NFS)/common internet file system (CIFS) protocol or a fiber channel (FC) protocol. The controller 00 includes a processor 001 and a cache 002. Specifically, the host 01 may send a data write request (referred to as a write request for short) to the controller 00. After the controller 00 receives the data write request, the data carried in the data write request is written into the hard disk 02. In addition, the host 01 can also send a data read request (referred to as a read request for short) to the controller 00. After the controller 00 receives the data read request, it searches for the data to be read in the buffer 002 according to the address in the data read request. If there is, directly send the data to be read (that is, the prediction file described later) to the host 01; if not, obtain the data from the hard disk 02 and send it to the host 01. In practical applications, the controller 00 and the aforementioned multiple hard disks 02 may be integrated in the storage device proposed in the embodiment of the present application, or may be used as two mutually independent devices to form the storage device proposed in the embodiment of the present application.

In the embodiment of the present application, a new function is configured for the processor 001, or one or more AI chips 003 are added to the controller 00, so that the processor 001 in the file prefetching method can The file that the host 01 accesses for the first time is predicted to implement prefetching of unvisited files, which is beneficial to the host 01 to increase the hit rate of read requests when accessing new files.

Based on the foregoing system architecture and application scenarios, the main process of the foregoing file prefetching method will be introduced below. As shown in Figure 1B, the storage device will perform the following steps:

101. Generate a first keyword according to the file access information carried in the read request.

In this embodiment, when a host (such as host 01 in FIG. 1A) needs to access a certain file in the storage medium, the host will issue a read request. Wherein, the read request is used to indicate the file to be accessed, and the read request will carry file access information of the file. The file access information refers to information related to the file that needs to be accessed. Optionally, the file access information may include file attribute information and access attribute information. Among them, the file attribute information is related to the file and is not affected by whether the file is accessed, for example, the file name, file type, file creator name, and access directory information. The access attribute information refers to the information involved in this access operation, and the access attribute information generated for different accesses to the same file is different. For example, the access time and the name of the file visitor. For ease of understanding, the aforementioned file access information is introduced by taking FIG. 2A as an example. As shown in Figure 2A, it is a folder containing four files. Take the file in the lower left corner as an example. The file name of the file is "2020 5th grade language simulation test paper 3"; the file type is text type (that is, the word type, which can also be called a word document); the file creator is "Zhang Teacher"; the access directory is "C:\Final Test Paper\Chinese", the access directory can also be called the storage path or storage address, which is used to indicate which folder the accessed file is stored in. In addition, information such as the access time and the name of the file visitor is not shown in FIG. 2A, and the information such as the access time and the name of the file visitor may be implicit in the aforementioned read request.

Optionally, the file access information is information of files from different access directories, that is, the read request may carry file access information for different files, and the aforementioned different files are from different access directories. Taking Figure 2A and Figure 2B as an example, the reading request is used to instruct to read the file whose access directory is "C:\Final Mock Paper\Language" and the file name is "2020 Fifth Grade Chinese Mock Paper 3" (specifically as Figure 2A), and the access directory is "C:\final mock test paper answer\language" and the file name is "2020 fifth grade language mock test answer 3" (specifically shown in Figure 2B).

At this time, the processor in the storage device (for example, the processor 001 in FIG. 1A) will obtain the file access information carried in the read request, and generate the first keyword based on the file access information. Wherein, the first keyword refers to a word formed by the storage device dividing or combining part of the file access information based on the aforementioned file access information. Optionally, the first keyword may be a word or a phrase, which is not specifically limited here. Optionally, the first keyword may be split from the file name. For example, the first keyword of a file whose file name is "2020 fifth grade Chinese simulation test paper 3" can be "2020" and "fifth grade". ", "Language mock test paper" or "3". Optionally, the first keyword may be formed by dividing the file name and then combining it, for example, "2020-May-3". Optionally, the first keyword may be a phrase composed of different file access information, for example, {"fifth grade Chinese", "Mr. Zhang", "word"}. In addition, the first keyword may also be a phrase composed of an access directory, for example, {"final mock test paper", "language"}. In practical applications, the aforementioned first keyword can be adjusted according to specific needs, and the specific embodiment of the present application does not limit the specific form of the first keyword.

102. Use the first keyword and the target access rule template to generate target keywords.

In this embodiment, after the storage device determines the first keyword, the storage device may generate the target keyword based on the aforementioned first keyword and the target access rule template. Wherein, the target keyword and the first keyword match the same characteristics, it can be understood that the thing indicated by the target keyword has the same characteristics as the thing indicated by the first keyword, and is based on the first keyword and The characteristics of the first keyword are easily associated with the target keyword. It can also be understood that the part of speech of the target keyword is the same as that of the first keyword, and the meaning of the first keyword is similar to or related to the target keyword. Optionally, the first keyword and the target keyword are both quantitative words, for example, the first keyword is "3" and the target keyword is "4"; for another example, the first keyword is "first section" , The target keyword is "Section 2". Optionally, the first keyword and the target keyword are synonyms or antonyms for each other. For example, the first keyword is "Book 1" and the target keyword is "Book 2"; for another example, the first keyword is "Exam Questions" ", the target keyword is "answer". The target keyword is used to indicate the prediction file. It can also be understood that the target keyword is included in the file attribute information of the prediction file. For example, the target keyword is the file name, file type, or file creator name of the prediction file, etc., which is not specifically limited here.

In addition, the target access rule template is an access rule template corresponding to the first keyword among the plurality of access rule templates. The target access rule template is used to input the first keyword and calculate the first keyword according to a certain rule to output the target keyword. The target access rule template can also be regarded as a comprehensive calculation model, and the target keyword can be output by inputting the first keyword.

103. Prefetch the prediction file indicated by the target keyword into the cache.

In this embodiment, after the storage device determines the target keyword, the storage device can use the target keyword to find the prediction file in a hard disk (for example, hard disk 02 in FIG. 1A), and prefetch the prediction file To the cache (for example, cache 002 in FIG. 1A). Optionally, the prefetched file may be a certain file or a certain group of related files, and the specifics are not limited here. For ease of understanding, the description will still be made by taking the aforementioned FIG. 2A and FIG. 2B as an example. Assume that the file name of the file to be accessed by the reading request is "2020 fifth grade language mock test paper 3", and the first keyword is the phrase {"final mock test paper", "language", "3"}. At this time, if predicted by the target access rule template, the target keyword is the phrase {"final simulation test paper answer", "language", "3"}. At this time, the storage device can determine that the file name of the prediction file is "2020 fifth grade Chinese simulation test paper answer 3". Then, the storage device can read the prediction file from the hard disk to the cache.

In this embodiment, the target keyword can be generated using the first keyword and the target access rule template, and then the prediction file to be prefetched is determined based on the target keyword. Since the first keyword and the target keyword match the same characteristics, it can be considered that the predicted file indicated by the target keyword is a file that the user needs to read at the next moment. Therefore, even if it is determined that the file access information of the first keyword comes from the file accessed for the first time, the storage device can use the aforementioned first keyword and the target access rule template to calculate the predicted file that has not been accessed, and calculate the predicted file Prefetch into the cache. It is helpful to improve the hit rate of the host acquiring data in the cache in the storage device.

Based on the foregoing embodiment, the foregoing file prefetching method is further introduced. Specifically, as shown in FIG. 3, the storage device will perform the following steps:

301a. Use a training model based on text semantics to train multiple initial keywords to obtain an access rule template.

In this embodiment, the storage device can pre-train the access rule template, so that when there is a read request, the storage device can generate the first keyword based on the file access information carried by the read request. Specifically, the storage device may use a training model based on text semantics to train a plurality of initial keywords to obtain an access rule template. The access rule template is generally stored in a cache in the storage device (for example, cache 002 in FIG. 1A). Optionally, the storage device may also back up the access rule template to a hard disk (for example, hard disk 02 in FIG. 1A) for subsequent recall at any time.

Among them, the training model based on text semantics means that the training model can recognize text semantics and perform association training based on text semantics. In practical applications, there are many types of training models based on text semantics, for example, a clustering model based on text semantics, which can classify keywords according to semantics. For another example, based on a neural network text semantic training model, the model may include the following functions: semantic textual similarity function, which is used to measure the basic semantics of text matching segments (for example, the aforementioned first keyword) Equivalence degree; Paraphrase identification (PI), that is, to identify whether two words (for example, the first keyword and target keyword mentioned above) express the same meaning; natural language inference, used to parse hypotheses The semantic similarity between the premise and the premise, etc. For another example, a naive Bayesian model based on text semantics can be trained based on the probability of multiple keywords appearing in the inference process.

The initial keyword is similar to the first keyword in the foregoing embodiment. The initial keyword can be generated by generating the first keyword. For example, the foregoing initial keyword is generated based on the relevant information carried by a certain historical read request or multiple times. word. For details, please refer to step 101 above. In addition, the initial keyword can also be preset by the user according to actual needs. Specifically, the initial keyword may be a word or a phrase, and may be a noun, an adjective, a quantifier, etc. The specific keyword is not limited here.

It should be understood that the storage device may execute the foregoing step 301a multiple times to generate multiple access rule templates, and the foregoing multiple access rule templates may form a set, which is referred to as an access rule template set in the embodiment of the present application. In addition, the storage device can also update each access rule template in the access rule template set at any time. Optionally, the access rule template set may be directly stored in the storage device as internal data of the storage device. In this case, the set of access rule templates contained in different storage devices may be different, and in addition, the same access rule template may also be different at different times. Optionally, the access rule template set can be integrated in a chip for the storage device to call. For example, the storage device may integrate the access rule template set into the AI chip. At this time, the AI chip (for example, AI chip 003 in FIG. 1A) may be located in a controller in a storage device (for example, in controller 00 in FIG. The AI chip makes a connection call.

301b. Generate a first keyword according to the file access information carried in the read request.

Wherein, the file access information is information related to the file indicated by the read request. The specific step 101 has been introduced in detail, and will not be repeated here.

Optionally, the storage device may also generate a second keyword based on the file access information carried in the read request. The second keyword is different from the first keyword, and the storage device will use the aforementioned first keyword and second keyword to perform subsequent prediction steps. In addition, the second keyword may also be a keyword from another document.

It should be understood that step 301a and step 301b are independent of each other. That is to say, when the storage device generates the first keyword based on the file access information, the storage device also uses the initial keyword to train the access rule template to generate the access rule template set. Therefore, there is no clear time sequence limitation between the foregoing step 301a and step 301b, and it can be understood that the storage device executes the foregoing step 301a and step 301b at the same time.

302. Use the first keyword to determine a target visit rule template in the set of visit rule templates.

In this embodiment, after the storage device determines the first keyword and the access rule template set, the storage device will search for the target visit corresponding to the first keyword in the access rule template set based on the first keyword. Regular template.

In this embodiment, the feature that each of the multiple initial keywords in the regular template matches each initial keyword is described as a feature related word. The characteristic associated word may be the common attribute of the aforementioned multiple initial keywords. For example, if a certain access rule template includes three initial keys of "3", "5" and "7", the characteristic related words of the aforementioned multiple initial keywords can be "odd", "positive integer", "prime number", etc. .

Since the target visit rule template is an visit rule template corresponding to the first keyword, the first keyword should conform to the characteristics indicated by the characteristic associated words in the visit rule template. Therefore, the target access rule template in this embodiment includes the first keyword and/or the first characteristic related words, where the first characteristic related words are used to indicate the characteristics that the first keyword conforms to. That is to say, the target access rule template may only include the first keyword, or may only include the first special related word, and may also include the aforementioned first keyword and the first characteristic related word. The details are not limited here.

Specifically, the storage device may search for the target access pattern template from the set of access pattern templates based on the first keyword in the following manner:

The storage device will traverse each access regularity template in the aforementioned access regularity template set. During the traversal process, the storage device will determine whether the access rule template contains the first keyword. If the access rule template includes the first keyword, the storage device determines that the access rule template is the target access rule template. Then, the storage device will skip the subsequent judgment and directly execute step 303. If the access rule template does not include the first keyword, the storage device will further determine whether the first keyword meets the characteristics indicated by the feature associated words in the access rule template. If the first keyword meets the characteristics indicated by the characteristic associated words in the visit rule template, then the visit rule template is determined to be the target visit rule template. Then, the storage device executes step 303.

Optionally, if the storage device generates two sets of keywords based on the file access information carried in the aforementioned read request, for example, the first keyword and the second keyword. At this time, the storage device needs to search for a target access rule template that matches both the second keyword and the first keyword in the set of access rule templates.

Specifically, the storage device may search for the target access rule template from the access rule template collection based on the first keyword and the second keyword in the following manner:

The storage device will traverse each access regularity template in the aforementioned access regularity template set. During the traversal process, the storage device will determine whether the access rule template contains the first keyword and the second keyword. If the visit rule template includes the first keyword and the second keyword, the storage device determines that the visit rule template is the target visit rule template. Then, the storage device will skip the subsequent judgment and directly execute step 303. If the access rule template only includes one of the first keyword and the second keyword or does not include the first keyword and the second keyword, the storage device further determines the first keyword and the second keyword. Whether the two keywords both conform to the characteristics indicated by the characteristic associated words in the access rule template. If the first keyword and the second keyword both conform to the characteristics indicated by the feature associated words in the access rule template, the storage device determines that the access rule template is the target access rule template. Then, the storage device executes step 303.

In addition, the target access rule template also includes a mapping relationship, and the mapping relationship is used to indicate an association manner between the plurality of initial keywords. When an initial keyword among the multiple initial keywords is used as the input of the mapping relationship, another keyword can be output, and the other keyword is also one of the foregoing initial keywords. Key words. The association method can be a function or a mapping table. For example, if the multiple initial keywords are "3", "5" and "7" respectively, the mapping relationship may be the function "y=x+2". For example, if the multiple initial keywords are "Language Test Paper Volume 1", "Language Test Paper Volume 2", and "Language Test Paper Volume 2", the mapping relationship may be a mapping table corresponding to the test paper and the answer.

303. Use the first keyword as the input of the mapping relationship, calculate the first keyword using the mapping relationship, and output the target keyword.

Wherein, the target keyword and the first keyword match the same characteristics. Specifically, the target keyword meets the characteristics of the first keyword indicated by the first characteristic associated word. The target keyword is used to indicate the prediction file. Specifically, the first keyword and the target keyword have been described in detail in step 102 above, and the details are not repeated here.

In this embodiment, since the target access rule template corresponds to the first keyword, the plurality of initial keywords may include the first keyword, or the first keyword may also be applicable to the foregoing mapping relationship. Therefore, the storage device can use the first keyword as the input of the mapping relationship, calculate the first keyword using the mapping relationship, and output the target keyword. For example, if the first keyword is "7" and the mapping relationship is the function "y=x+2", then the target keyword "9" is output. For example, if the multiple initial keywords are "Language Test Paper Volume 1", "Language Test Paper Volume 2", and "Language Test Paper Volume 2", the mapping relationship may be a mapping table corresponding to the test paper and the answer. If you enter "Language Test Paper Volume 1", it will output "Language Test Paper Answer Volume 1". In practical applications, there are still many examples, and the details are not repeated here.

304. Prefetch the prediction file indicated by the target keyword into the cache.

In this embodiment, step 304 is similar to the aforementioned step 103. For details, please refer to step 103, which will not be repeated here.

In this embodiment, the first keyword is used as the input of the mapping relationship in the target access rule template to output the target keyword, and then the prediction file to be prefetched is determined based on the target keyword. Since the first keyword and the target keyword match the same characteristics, it can be considered that the predicted file indicated by the target keyword is a file that the user needs to read at the next moment. Therefore, even if it is determined that the file access information of the first keyword comes from the file accessed for the first time, the storage device can use the aforementioned first keyword and the target access rule template to calculate the predicted file that has not been accessed, and calculate the predicted file Prefetch into the cache. It is helpful to improve the hit rate of the host acquiring data in the cache in the storage device.

Optionally, in the foregoing step 301a, the storage device is always in a state of updating and maintaining the access rule template set. When a target access rule template corresponding to the first keyword is found based on the first keyword, the storage device will combine the target access rule template and other access rule templates in the access rule template set to the access rule template. The collection is updated and maintained. For ease of introduction, it is said that the access rule template set includes one target access rule template and at least one candidate access rule template. Specifically, the storage device determines the degree of association between the target access rule template and each candidate access rule template in at least one candidate access rule template, where the association degree is used to indicate the mapping relationship between the target access rule template and the candidate access rule template. The degree of similarity between the mapping relationships in the candidate access pattern template. Then, the storage device merges the candidate visit rule template corresponding to the degree of association higher than the preset value with the target candidate visit rule template to obtain an updated set of visit rule templates.

In this embodiment, it is proposed that the access rule template set in the storage device can be updated based on the target access rule template corresponding to the first keyword, that is, the storage device updates the predicted file based on the first keyword. A collection of access rule templates in the storage device. Such an implementation manner can enable each access rule template in the set of multiple updated access rule templates to more accurately determine a more accurate prediction file based on the input keywords, so that the aforementioned prediction file can be prefetched to the cache. Improve the hit rate of read requests sent by the host.

The structure of the storage device in the foregoing embodiment will be introduced below:

As shown in FIG. 4, a schematic structural diagram of a storage device 40 provided in this embodiment. The storage device in the method embodiment corresponding to FIG. 1B and FIG. 3 may be based on the structure shown in FIG. 4 in this embodiment.

The storage device 40 includes at least one processor 401 and at least one storage medium 402.

The processor 401 may be a general-purpose central processing unit (central processing unit, CPU) or a microprocessor (microprocessor). The processor 401 may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). In addition, the processor 401 may refer to one or more devices, circuits, and/or processing cores for processing data (for example, computer program instructions). In this embodiment, the processor 401 may be configured to receive a read request issued by the host, and generate keywords (such as the first keyword and the second keyword, etc.) based on the file access information carried in the read request. In addition, the processor 401 is also configured to use the first keyword and the target access rule template to generate a target keyword, and prefetch a prediction file based on the target keyword. Specifically, the processor 401 may execute other steps in the aforementioned embodiment corresponding to FIG. 1B and FIG. 3.

In particular, when the storage device 40 executes step 301a in the embodiment corresponding to FIG. 3, any one of the following implementation manners can be adopted:

Optionally, the processor 401 is further configured to perform training based on a plurality of initial keywords to generate an access rule template, and update the set of access rule templates based on the aforementioned target access rule template.

Optionally, the aforementioned at least one processor 401 includes one or more first processors 4011, and the first processors 4011 are configured to perform training based on a plurality of initial keywords to generate an access rule template, and based on the aforementioned target access rule The template updates the access rule template collection. The first processor 4011 may be a functional module or an independent functional chip in the processor 401, and is used to maintain the aforementioned access rule template set. The first processor 4011 may also be an AI chip with specific computing functions.

Optionally, the storage device 40 further includes an AI chip 403. The AI chip 403 is located outside the processor 401 and is used to implement the function of the aforementioned first processor 4011, that is, the AI chip 403 is used to base multiple initial keywords. Training is performed to generate an access rule template, and the access rule template set is updated based on the aforementioned target access rule template. At this time, the aforementioned processor 401 and the AI chip 403 also include a chip interface, and the processor 401 communicates with the AI chip 403 through the chip interface, so as to call the access rule template set generated in the AI chip 403.

In the foregoing three optional implementation manners, for the steps of generating access rule templates and updating and maintaining the access rule template set, please refer to the related introduction in step 301b in the embodiment corresponding to FIG. 3 for details. Go into details again.

In addition, the storage medium 402 includes a cache memory 4021 and a hard disk 4022. Wherein, the cache 4021 may also be referred to as a memory, which is a bridge for communication between external storage (that is, the hard disk 4022) and the processor 401. The cache 4021 can be used to temporarily store the arithmetic data in the processor 401 and data exchanged with an external memory such as the hard disk 4022. When the computer is running, the processor 401 can transfer the data to be calculated from the cache 4021 to the processor 401 for calculation. It should be understood that the hard disk 4022 and the cache 4021 further include one or more interfaces, and the one or more interfaces are used to implement data transmission between the hard disk 4022 and the cache 4021. Specifically, the processor 401 is configured to prefetch the prediction file from the hard disk 4022 to the cache 4021 based on the target keyword.

In addition, the cache 4021 is also used to store the access rule template generated by the aforementioned processor 401/first processor 4011/AI chip 403. The processor 401/first processor 4011/AI chip 403 can also back up the aforementioned access rule template to the hard disk 4022, so that the processor 401/first processor 4011/AI chip 403 can access the The access rule template is called in the hard disk 4022. In particular, since the AI chip 403 also contains a part of storage media, when the AI chip 403 generates the access rule template, the access rule template can be directly stored in the AI chip 403.

Optionally, the aforementioned processor 401, cache 4021, and AI chip 403 are generally located in the same controller (not shown). Specifically, you can refer to the system architecture diagram corresponding to FIG. 1A, and the details are not repeated here.

As shown in FIG. 5, it is a schematic structural diagram of a prefetching device 50 provided by an embodiment of this application. The prefetching device 50 is located in the aforementioned storage device 40. For the specific structure of the storage device 40, please refer to the aforementioned FIG. 4. The storage device 40 stores a computer program or instruction, and the prefetching device 50 invokes the computer program or instruction to execute the following modules: a keyword generation module 501, a calculation module 502, a data migration module 503, and a template generation module 504. The keyword generation module 501, the calculation module 502, and the data migration module 503 are located in the processor 401 in the storage device 40 shown in FIG. 4; the template generation module 504 may be located in the storage device 40 shown in FIG. The processor 401 is, for example, the first processor 4011 in the processor 401; the template generation module 504 may also be located in the AI chip 403 in the storage device 40 shown in FIG. 4.

Among them, the keyword generating module 501 is configured to generate the first keyword according to the file access information carried in the read request. Specifically, please refer to the relevant introduction in the foregoing step 101 and the foregoing step 301b. The calculation module 502 is configured to use the first keyword and the target access rule template to generate a target keyword. The target keyword matches the same characteristics as the first keyword, and the target keyword is used to indicate a prediction file. Specifically, reference may be made to the related introductions in the foregoing step 102, the foregoing step 302, and the foregoing step 303. The data migration module 503 is configured to prefetch the prediction file indicated by the target keyword from the hard disk to the cache. Specifically, reference may be made to the related introduction of the foregoing step 103 and the foregoing step 304.

Wherein, the prefetching device 50 stores multiple access rule templates, or the prefetching device 50 can call multiple access rule templates stored in the storage device. The target access rule template is an access rule template corresponding to the first keyword among the plurality of access rule templates, and each of the access rule templates is obtained by training a plurality of initial keywords using a training model based on text semantics. The target visit rule template includes the first keyword and/or first characteristic related words, and the first characteristic related words are used to indicate the characteristics that the first keyword conforms to. The target access rule template further includes a mapping relationship, and the mapping relationship is used to indicate an association manner between the plurality of initial keywords, and the plurality of initial keywords include the first keyword.

In addition, the calculation module 502 is specifically configured to use the first keyword as the input of the mapping relationship, calculate the first keyword using the mapping relationship, and output the target keyword. The target keyword matches the first keyword. The feature of the first keyword indicated by the feature related word. Optionally, the calculation module 502 is further configured to: determine whether the access rule template includes the first keyword; if the access rule template includes the first keyword, determine that the access rule template is the target access rule template; If the access rule template does not contain the first keyword, then it is further determined whether the first keyword meets the characteristics indicated by the feature related words in the access rule template; if the first keyword meets the feature related words in the access rule template Indicates the characteristic, the visit rule template is determined to be the target visit rule template. Specifically, please refer to the relevant introduction in 303 above.

In addition, the template generation module 504 is used to train the plurality of initial keywords using the text semantic training model to obtain the access rule template. Specifically, reference may be made to the related introduction of the foregoing step 301a.

One or more of the above modules can be implemented by software, hardware or a combination of both. When any of the above modules is implemented by software, the software exists in the form of computer program instructions and is stored in the memory, and the processor can be used to execute the program instructions and implement the above method flow. The processor at this time may include but is not limited to at least one of the following: central processing unit, microprocessor, digital signal processing (digital signal processing, DSP), microcontroller (microcontroller unit, MCU), or artificial intelligence processing Various computing devices that run software, such as a computer, and each computing device may include one or more cores for executing software instructions to perform operations or processing. The processor may be built in a system on chip (SoC) or an application specific integrated circuit (ASIC), or it may be an independent semiconductor chip. The processor's internal processing is used to execute software instructions to perform calculations or processing, and may further include necessary hardware accelerators, such as field programmable gate array (FPGA) and programmable logic circuit (programmable logic). device, PLD) or a logic circuit that implements dedicated logic operations. When the above modules or units are implemented in hardware, the hardware can be CPU, microcontroller, DSP, MCU, artificial intelligence processor, ASIC, SoC, FPGA, PLD, dedicated digital circuit, hardware accelerator or non-integrated discrete device For any one or any combination of the above, it can run necessary software or does not rely on software to perform the above method flow.

Those skilled in the art can clearly understand that, for the convenience and conciseness of the description, the specific working process of the above-described system, device, and unit can refer to the corresponding process in the foregoing method embodiment, which will not be repeated here.

The above embodiments are only used to illustrate the technical solutions of the present application, not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still compare the previous embodiments. The recorded technical solutions are modified, or some of the technical features are equivalently replaced; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

A file prefetching method, applied to a storage device, characterized in that it includes:

Generating a first keyword according to the file access information carried in the read request, where the file access information is information related to the file indicated by the read request;

Generating a target keyword using the first keyword and a target access rule template, the target keyword conforms to the same characteristics as the first keyword, and the target keyword is used to indicate a prediction file;

Prefetch the prediction file indicated by the target keyword into a cache.
The method according to claim 1, wherein the storage device stores a plurality of access rule templates, and the target access rule template is an access corresponding to the first keyword among the plurality of access rule templates. Regular template, each of the access regular templates is obtained by training a plurality of initial keywords using a training model based on text semantics.
The method according to claim 2, wherein the target access rule template contains the first keyword and/or first characteristic related words, and the first characteristic related words are used to indicate that the first keyword matches Characteristics.
The method according to claim 3, wherein the target access rule template further comprises a mapping relationship, and the mapping relationship is used to indicate an association manner between the plurality of initial keywords, and the plurality of initial keywords The word includes the first keyword;

The generating target keywords using the first keyword and the target access rule template includes:

The first keyword is used as the input of the mapping relationship, the first keyword is calculated using the mapping relationship, and the target keyword is output, and the target keyword meets the first characteristic associated word indication The characteristics of the first keyword.
The method according to claim 3 or 4, wherein the method further comprises:

Judging whether the access rule template contains the first keyword;

If the access rule template includes the first keyword, determining that the access rule template is the target access rule template;

If the access pattern template does not include the first keyword, it is further determined whether the first keyword meets the characteristics indicated by the characteristic associated words in the access pattern template;

If the first keyword meets the characteristics indicated by the characteristic associated words in the visit rule template, it is determined that the visit rule template is the target visit rule template.
The method according to any one of claims 1 to 4, wherein the file access information is information of files from different access directories, and the multiple predicted files are located in different access directories.
A storage device, characterized by comprising: a cache, a hard disk, and at least one processor;

The at least one processor is used to:

Generating a first keyword according to the file access information carried in the read request, where the file access information is information related to the file indicated by the read request;

Generating a target keyword using the first keyword and a target access rule template, the target keyword conforms to the same characteristics as the first keyword, and the target keyword is used to indicate a prediction file;

Prefetching the prediction file indicated by the target keyword from the hard disk into the cache.
The storage device according to claim 7, wherein the storage device stores a plurality of access rule templates, and the target access rule template is one of the plurality of access rule templates corresponding to the first keyword Access regularity templates, each of the access regularity templates is obtained by training a plurality of initial keywords using a training model based on text semantics.
The storage device according to claim 8, wherein the target access rule template contains the first keyword and/or first characteristic related words, and the first characteristic related words are used to indicate the first keyword Compatible characteristics.
The storage device according to claim 9, wherein the target access rule template further comprises a mapping relationship, and the mapping relationship is used to indicate an association manner between the plurality of initial keywords, and the plurality of initial keywords Keywords include the first keyword;

The at least one processor is specifically configured to use the first keyword as the input of the mapping relationship, calculate the first keyword using the mapping relationship, and output the target keyword, the target The keyword conforms to the feature of the first keyword indicated by the first feature related word.
The storage device according to claim 9 or 10, wherein the at least one processor is further configured to:

Judging whether the access rule template contains the first keyword;

If the access rule template includes the first keyword, determining that the access rule template is the target access rule template;

If the access pattern template does not include the first keyword, it is further determined whether the first keyword meets the characteristics indicated by the characteristic associated words in the access pattern template;

If the first keyword meets the characteristics indicated by the characteristic associated words in the visit rule template, it is determined that the visit rule template is the target visit rule template.
The storage device according to any one of claims 7 to 10, wherein the file access information is information of files from different access directories, and the multiple predicted files are located in different access directories.
The storage device according to any one of claims 7 to 10, wherein the at least one processor further comprises a first processor, and the first processor is configured to use the text-based semantic training model pair The multiple initial keywords are trained to obtain the access rule template.
A prefetching device, characterized in that the prefetching device is located in a storage device, and the storage device further includes a cache and a hard disk;

The storage device stores a computer program or instruction, and the prefetching device invokes the computer program or instruction to execute the following modules:

The keyword generation module is configured to generate a first keyword according to the file access information carried in the read request, where the file access information is information related to the file indicated by the read request;

A calculation module, configured to use the first keyword and the target access rule template to generate a target keyword, the target keyword and the first keyword have the same characteristics, and the target keyword is used to indicate a prediction file;

The data migration module is configured to prefetch the prediction file indicated by the target keyword from the hard disk to the cache.
The prefetching device of claim 14, wherein the prefetching device stores a plurality of access rule templates, and the target access rule template is the first keyword among the plurality of access rule templates. Corresponding access rule templates, each of the access rule templates is obtained by training a plurality of initial keywords using a training model based on text semantics.
The prefetching device according to claim 15, wherein the target access rule template contains the first keyword and/or first characteristic related words, and the first characteristic related words are used to indicate the first key The characteristics of the word match.
The prefetching device according to claim 16, wherein the target access rule template further comprises a mapping relationship, and the mapping relationship is used to indicate an association manner between the plurality of initial keywords, and the plurality of The initial keywords include the first keyword;

The calculation module is specifically configured to use the first keyword as the input of the mapping relationship, calculate the first keyword using the mapping relationship, and output the target keyword, the target keyword Meet the feature of the first keyword indicated by the first feature related word.
The prefetching device according to claim 16 or 17, wherein the calculation module is further configured to:

Judging whether the access rule template contains the first keyword;

If the access rule template includes the first keyword, determining that the access rule template is the target access rule template;

If the access pattern template does not include the first keyword, it is further determined whether the first keyword meets the characteristics indicated by the characteristic associated words in the access pattern template;

If the first keyword meets the characteristics indicated by the characteristic associated words in the visit rule template, it is determined that the visit rule template is the target visit rule template.
The prefetching device according to any one of claims 14 to 17, wherein the file access information is information of files from different access directories, and the multiple predicted files are located in different access directories.
The prefetching device according to any one of claims 14 to 17, wherein the prefetching device further comprises a template generation module; the template generation module is configured to use the text semantic-based training model pair The multiple initial keywords are trained to obtain the access rule template.