CN102981963B - A kind of implementation method of flash translation layer (FTL) of solid-state disk - Google Patents

Abstract

The invention discloses a method for realizing a flash conversion layer of a solid-state disk, comprising: receiving a read-write request sent by a file system, the read-write request corresponds to a page with a logical address K, and judging whether the logical address K is in a cache mapping table Hit, if not, then judge whether the cache mapping table can accommodate the number n of mapping relations transferred into the cache mapping table next time, if not, find out a least recently used one in the cache mapping table The logical page number Victim_Num of a mapping relationship, find out the mapping relationship with Victim_Num in the cache mapping table, store all logical page numbers in the same page in the mapping table, and find out from the mapping relationship corresponding to these logical page numbers For the n least recently used mappings, find the corresponding mappings in the global conversion directory according to Victim_Num. The invention has higher reading and writing efficiency, improves the performance of the system, and prolongs the service life of the solid state disk.

Description

A method for implementing a flash conversion layer of a solid state disk

technical field

The invention belongs to the technical field of solid-state disk storage, and more particularly relates to a method for realizing a flash conversion layer of a solid-state disk.

Background technique

In recent years, flash memory (Flash) technology has led new changes in the storage field. With its small size, fast speed, low power consumption, no noise and anti-vibration and many other advantages, it has gradually replaced the traditional mechanical hard disk.

Completely different from traditional hard disks, as shown in Figure 1, a flash memory chip (FlashChip) 101 in the prior art is composed of multiple blocks (Block) 102, and one block 102 is composed of multiple pages (Page) 103, usually 64 , and a page 103 is divided into two areas, which are composed of a data area 104 (DataArea) and an OOB area (Out-ofBandarea) 105 respectively. Generally, the data area in a page 103 is 512B, and the OOB is 16B. Among them, the main function of OOB is to record the Hamming check code, the logical page number and the status of the current page, that is, Free/Valid/Invalid (Free/Valid/Invalid).

Based on the structural characteristics of flash memory itself, it also has its disadvantages, which are mainly manifested in four aspects: 1. When reading and writing, use the page as the unit, not the data; 2. Before writing, use the block Erase as a unit, that is, it cannot overwrite (Over-write). When a piece of data needs to be modified, it cannot be directly modified in situ like a disk, but the data block needs to be erased and then rewritten into the data; 3. The erasing times of the Block are limited, that is, the lifetime of the flash memory is limited. Therefore, flash memory cannot be directly used by traditional file systems like mechanical disk storage.

In order to make full use of the technologies and products accumulated in the field of traditional disks, major manufacturers have combined and packaged one or more flash memory chips into a solid state drive (SSD) similar to a disk, providing upper-layer applications with the same interface as traditional disks. Without the need to modify the file system and applications. Therefore, a flash translation layer (Flash translation layer, referred to as FTL) is needed to simulate the solid-state disk into a standard block device to shield its characteristics, so that the upper-level file system uses it like an ordinary disk storage.

As shown in Fig. 2, the solid-state disk system of the prior art includes file system (in sector unit), FTL, flash drive and flash memory chip from top to bottom, and wherein FTL includes three big blocks again, and address mapping is about upper file system The logical address in units of sectors is converted into the physical address of the solid state disk, which is the most important part of FTL. Garbage collection and wear leveling enable each flash memory to recycle failed flash memory blocks under wear leveling conditions.

In the existing FTL, although the feature of erasing before writing can be shielded from the upper system, there are still obvious deficiencies: 1. The cache scheduling algorithm in it is not well combined with the read and write characteristics of the solid state disk; For example, the commonly used page-level mapping FTL now uses the LRU algorithm commonly used in the traditional file system used in the transfer of the mapping relationship to the cache, that is, only one mapping relationship is transferred at a time, but the SSD uses pages as read and write units. Calling in a mapping relationship requires reading the entire mapping page, and calling out to update a mapping relationship requires writing the entire mapping page. It shows that this cache management mechanism is not suitable for solid-state disks; 2. The cache scheduling in the existing FTL utilizes the spatial locality of data, but the utilization of sequence locality is not yet mature. There may be multiple layers of cache in an actual storage system, and the cache of the solid-state disk is at the bottom layer. When the I/O requests of the system reach the cache of the solid-state disk after being filtered by the caches of the upper layers, there is almost no spatial locality. Therefore, the hit rate of the existing FTL call-out algorithm applied to the solid-state disk is not ideal. The SSD cache miss overhead is very high (if the mapping needs to be updated, the miss overhead includes two read operations and one write operation), which seriously reduces the life of the hard disk.

Contents of the invention

Aiming at the defects of the prior art, the object of the present invention is to provide a method for implementing the flash memory conversion layer of a solid-state disk, which has higher read and write efficiency and is more suitable for solid-state disks, and effectively improves the performance through multi-table forward scheduling and batch update Improve the cache hit rate and reduce the update times of the flash memory, thereby improving the performance of the system and prolonging the life of the solid-state disk.

In order to achieve the above object, the present invention provides a method for implementing a flash conversion layer of a solid-state disk, comprising the following steps:

(1) Receive the read-write request sent by the file system, the read-write request corresponds to the page whose logical address is K, where K is a positive integer;

(2) judge whether logical address K hits in the cache mapping table, if it is then enter step (13), otherwise enter step (3);

(3) judge whether the mapping relation number n that can be transferred into the cache mapping table next time can be accommodated in the cache mapping table again, if it can accommodate, then turn to step (11), otherwise turn to step (4);

(4) Find out a logical page number Victim_Num of a least recently used mapping relationship in the cache mapping table;

(5) Find out the mapping relationship with Victim_Num in the cache mapping table and store all logical page numbers in the same page in the mapping table, and find n least recently used mappings from the mapping relationship corresponding to these logical page numbers relation;

(6) Find the corresponding mapping relationship in the global conversion directory according to Victim_Num. Specifically, after dividing 512 by Victim_Num, it is E, and the remainder is F, indicating that the mapping relationship with the logical page number Victim_Num is stored in the logical mapping page number E , the offset is F, and the page whose logical mapping page number is E is stored in the location where the physical mapping page number is B through the global conversion directory;

(7) Find the page whose physical mapping page number is B in the mapping table, and compare it with the mapping relationship in the cache mapping table to determine whether there is a logical page number that is the same but the corresponding physical page number is different. If so, go to step (8); otherwise go to step (10);

(8) Find an available page in the mapping table, whose physical mapping page number is C, update the mapping relationship that the physical mapping page number B appears in the cache mapping table to this available page C, and put the physical mapping page number B Copy the mapping relationship that does not appear in the cache mapping table to this available page C, set the page with the physical mapping page number B as invalid, wait for recycling, and set the page with the physical mapping page number C from available to valid ;

(9) Change the physical mapping page number B corresponding to the logical mapping page number E to C in the global conversion directory;

(10) remove n least recently used page mapping relationships in the cache mapping table;

(11) After determining that the logical address K is divided by 512, it is G, and the remainder is H, and finds the physical mapping page number M corresponding to the logical mapping page number G in the global translation directory;

(12) Find the page whose physical mapping page number is M in the mapping table, and then find the mapping relationship of its offset H, call it and the following consecutive n-1 mapping relationships in the same page, and add it to the cache in the mapping table;

(13) Find the record whose logical page number is K in the cache mapping table, and obtain its physical page number J;

(14) Find the page whose physical page number is J in the data block, and return the address of this page to the file system.

The value of the number n of mapping relations transferred into the cache mapping table at one time is 4-16.

Through the above technical solutions conceived by the present invention, compared with the prior art, the present invention has the following beneficial effects:

1. The cache hit rate is improved: because in step (13), multi-table forward scheduling is carried out, and multiple consecutive mapping relationships are transferred into one read operation, and the sequential locality of the underlying cache is fully utilized.

2. Reduce the number of reads and writes of the flash memory: because the batch update technology is adopted in steps (5) and (10), batch updates can be performed without increasing the number of reads and writes.

3. The life-span of the flash memory is extended: because the steps (5) and (10) are adopted, the number of reads and writes of the Flash memory is reduced, thereby prolonging its service life.

4. Sequence locality is utilized: because step (13) is adopted, the mapping relationship transferred into the CMT is consecutive n records in the same page. At the same time as the currently required mapping is loaded, the mapping that is likely to be used in the future is loaded, not only using the spatial locality (in the upper cache), but also effectively using the sequential locality, even after the filtering of the upper layer I/O It can further greatly improve the cache hit ratio.

Description of drawings

FIG. 1 is a schematic structural diagram of a flash memory chip in the prior art.

FIG. 2 is a schematic structural diagram of a solid state disk system in the prior art.

Fig. 3 is a design structure diagram of the implementation method of the flash conversion layer of the solid state disk of the present invention.

Fig. 4 is a schematic diagram of the flow of reading and writing in the method of the present invention.

Fig. 5 is a flow chart of the implementation method of the flash conversion layer of the solid state disk of the present invention.

FIG. 6 is a comparison of the read and write times between the flash memory conversion method of the prior art and the method of the present invention.

FIG. 7 is a comparison of the erasing times between the flash conversion method of the prior art and the method of the present invention.

FIG. 8 is a comparison of the cache hit rate between the flash conversion method of the prior art and the method of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

The technical terms of the present invention are firstly explained and illustrated below.

Cache mapping table: CacheMappingTable, hereinafter referred to as CMT, is used to store the active mapping table;

Mapping table: TranslationBlocks, which stores the mapping relationship between all logical page numbers and physical page numbers;

Data block: DataBlocks, used to store the user's real data;

Global translation directory: GlobalTranslationDirectory, which is used to track which page in the mapping table the mapping relationship of the logical address number exists;

D _LPN : LogicalDataPageNumber, logical page number;

D _PPN : PhysicalDataPageNumber, physical page number;

M _VPN : VirtualTranslationPageNumber, logical mapping page number;

M _PPN :: PhysicalTranslationPageNumber, physical mapping page number;

The technical solutions in the examples of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the implementation examples of the present invention. Apparently, the described embodiments are some examples of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts fall within the protection scope of the present invention.

As shown in Figure 3, the flash memory chip that the method for implementing the flash conversion layer of the solid-state disk of the present invention is applied is divided into four large blocks altogether, and the cache mapping table (CachedMappingTable, referred to as CMT) is used for storing the mapping table in the activity; The translation directory (GlobalTranslationDirectory, referred to as GTD), is used to track which page of the mapping table the mapping relationship of the logical address number exists in; both CMT and GTD are stored in the SRAM; the data block (DataBlock) and the mapping table (TranslationBlock), which are located in In the flash memory, the data block accounts for the vast majority, and the data block stores real user data, while the mapping table stores the mapping relationship from all logical addresses in the data block to physical addresses.

The present invention fully considers the read-write characteristics of the solid-state disk. When the cache mapping table is full and needs to call out a part of the mapping relationship, it will update all the other mapping relationships in the mapping table and this record in the same page (referred to as Batch update technology), so that without increasing the number of reads and writes, the number of writes to the mapping table is reduced, thereby prolonging the life of the SSD.

In addition, in the present invention, the combination with the upper layer cache is fully considered in the call-in strategy. Since the spatial locality has been fully utilized in the upper-level cache, considering the filtering effect of the upper-level cache, the LRU hit rate in the solid-state disk will be greatly reduced, so the present invention utilizes sequential locality in the call-in strategy, that is, multiple The continuous mapping relationship is in the cache mapping table (referred to as multi-table forward scheduling). Experiments show that this greatly improves the cache hit rate.

In addition, when the cache mapping table is full and multiple mapping relationships need to be called in, it is necessary to call out multiple mapping relationships in the cache. At this time, first find out the logical page number Victim_Num of the least recently used mapping relationship in the CMT, then find out all the mapping relationships that appear in the CMT on the same page as Victim_Num in the mapping table, and then find out n-1( Where n is the number of mapping relationships transferred into the CMT once) and the least recently used mapping relationships are called out.

As shown in Figures 4 and 5, the implementation method of the flash conversion layer of the solid state disk of the present invention comprises the following steps:

(1) Receive the read-write request sent by the file system, the read-write request corresponds to the page whose logical address is K (K is a positive integer), as shown in (1) in Figure 4, in the illustrated example, K= 1280;

(2) judge whether logic address K hits in CMT, if then enter step (13), otherwise enter step (3);

(3) Judging whether n can be accommodated in the CMT, where n is the number of mapping relationships transferred into the CMT at one time, if it can be accommodated, then go to step (11), otherwise go to step (4); usually n is taken The value is 4-16, in the illustrated example, n=4;

(4) find out the logical page number Victim_Num of a mapping relation that uses least recently in CMT, as shown in (4) among Fig. 4, in the example shown in figure, get VictimNum=1;

(5) Find out the mapping relationship with Victim_Num in CMT and store all logical page numbers in the same page in the mapping table (because the logical page numbers in the mapping table are sequential, such as logical numbers 0-511 are on the same page , 512-1024 on the same page), and find n least recently used mapping relationships from the mapping relationships corresponding to these logical page numbers, as shown in (5) in Figure 4, n logical page numbers are respectively 1, 2, 4, 7;

(6) Find the corresponding mapping relationship in GTD according to Victim_Num (in GTD, the mapping relationship records are stored sequentially). Specifically, after Victim_Num is divided by 512 in the figure, it is E, and the remainder is F, indicating the logical page number The mapping relationship of Victim_Num is stored in the position where the logical mapping page number is E and the offset is F, and the page whose logical mapping page number is found through GTD is stored in the position where the physical mapping page number is B; for example, as shown in As shown in (7) in 4, B=21, Victim_Num is 0 after being divided by 512, and the remainder is 1, indicating that the mapping relationship with the logical page number Victim_Num is stored at the position where the logical mapping page number is 0 and the offset is 1, through GTD Find the page whose logical mapping page number is 0 and store it in the location where the physical mapping page number is 21;

(7) Find the page whose physical mapping page number is B in the mapping table, and compare it with the mapping relationship in the CMT, and judge whether there is a logical page number that is the same but the corresponding physical page number is different, if so, go to step (8 ); otherwise go to step (10);

(8) Find an available page in the mapping table, whose physical mapping page number is C, update the mapping relationship that the physical mapping page number B appears in the CMT to this available page C, and update the physical mapping page number B that does not have The mapping relationship that appears in the CMT is copied to the available page C, and the page with the physical mapping page number B is set to be invalid, waiting for recycling, and the page with the physical mapping page number C is set from available to valid, as shown in Figure 4 Shown in (8), in the illustrated example, get C=23;

(9) Change the physical mapping page number B corresponding to the logical mapping page number E to C in GTD, for example, as shown in (9) in Figure 4, the physical mapping corresponding to the logical mapping page number 0 in GTD Page number 21 changed to 23;

(10) remove n least recently used page mapping relationships in CMT, in the illustrated example, be four records that logical page numbers are 1, 2, 4, 10;

(11) Determine the logical address K divided by 512 is G, and the remainder is H, and find the physical mapped page number M corresponding to the logical mapped page number G in GTD, as shown in (11) in Figure 4, the logical address 1280 divided by 512 Back G=2, remainder H=256, M=15;

(12) Find the page whose physical mapping page number is M in the mapping table, and then find the mapping relationship of its offset H, call it and the following consecutive n-1 mapping relationships in the same page, and add it to the CMT In, as shown in (12) among Fig. 4;

(13) find the record that logical page number is K in CMT, obtain its physical page number J, as shown in (13) among Fig. 4, J=660;

(14) Find the page whose physical page number is J in the data block, and return the address of this page to the file system.

Fig. 6 and Fig. 7 are respectively the contrast of reading and writing performance and erasing times of existing flash memory conversion method and the method of the present invention, as can be seen from the figure, the method of the present invention can effectively reduce the times of reading and writing and erasing times, Among them, the number of reads is less by 54.21%, the number of writes is less by 15.74%, and the number of erasures is less by 15.75%.

FIG. 8 is a comparison of the cache hit rate between the existing flash memory conversion method and the method of the present invention. It can be seen from the figure that the cache hit rate has increased by 20.1%.

It is easy for those skilled in the art to understand that the above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention, All should be included within the protection scope of the present invention.

Claims (2)

1. an implementation method of the flash conversion layer of solid-state disk, is characterized in that, comprises the following steps: (1) Receive the read-write request sent by the file system, the read-write request corresponds to the page whose logical address is K, where K is a positive integer; (2) judge whether logical address K hits in the cache mapping table, if it is then enter step (13), otherwise enter step (3); (3) judge whether the mapping relation number n that can be transferred into the cache mapping table next time can be accommodated in the cache mapping table again, if it can accommodate, then turn to step (11), otherwise turn to step (4); (4) Find out a logical page number Victim_Num of a least recently used mapping relationship in the cache mapping table; (5) Find out the mapping relationship with the logical page number Victim_Num in the cache mapping table. Store all logical page numbers in the same page in the mapping table, and find out n least recent from the mapping relationship corresponding to these logical page numbers The mapping relationship used; (6) Find the corresponding mapping relationship in the global conversion directory according to the logical page number Victim_Num. Specifically, after dividing 512 by the logical page number Victim_Num, it is E, and the remainder is F, indicating that the logical page number is the mapping of the logical page number Victim_Num The relationship is stored at the position where the logical mapping page number is E and the offset is F, and the page with the logical mapping page number E found through the global conversion directory is stored at the position where the physical mapping page number is B; (7) Find the page whose physical mapping page number is B in the mapping table, and compare it with the mapping relationship in the cache mapping table to determine whether there is a logical page number that is the same but the corresponding physical page number is different. If so, go to step (8); otherwise go to step (10); (8) Find an available page in the mapping table, whose physical mapping page number is C, update the mapping relationship that the physical mapping page number B appears in the cache mapping table to this available page C, and put the physical mapping page number B Copy the mapping relationship that does not appear in the cache mapping table to this available page C, set the page with the physical mapping page number B as invalid, wait for recycling, and set the page with the physical mapping page number C from available to valid ; (9) Change the physical mapping page number B corresponding to the logical mapping page number E to C in the global conversion directory; (10) remove n least recently used page mapping relationships in the cache mapping table; (11) After determining that the logical address K is divided by 512, it is G, and the remainder is H, and finds the physical mapping page number M corresponding to the logical mapping page number G in the global conversion directory; (12) Find the page whose physical mapping page number is M in the mapping table, and then find the mapping relationship of its offset H, call it and the following consecutive n-1 mapping relationships in the same page, and add it to the cache in the mapping table; (13) Find the record whose logical page number is K in the cache mapping table, and obtain its physical page number J; (14) Find the page whose physical page number is J in the data block, and return the address of this page to the file system. 2. The implementation method according to claim 1, characterized in that, the value of the number n of mapping relationships transferred into the cache mapping table at one time is 4-16.