RU2636107C1 - Method of recording data to digital information drive on basis of nand type flash-memory - Google Patents

Abstract

FIELD: information technology.

SUBSTANCE: at the beginning of the user's session with the drive, the user's script is automatically selected from the set of suggestions. Logging is performed only for the most frequently overwritten pages, determined by the script selected at the beginning of the user's session with the drive. The merge of data block and the corresponding log block is performed when a request is received to read the logging page from this block. For operations of pages overwriting that are not related to logged, a "deferred write" of blocks is performed, which consists of writing information to the buffer block instead of the address block until the current address block number is changed or the first data reading request.

EFFECT: extending life of flash memory and increasing the speed of recording information on the media.

7 dwg

Description

FIELD OF THE INVENTION

The invention relates to computer technology and is a method of writing data to a digital information storage device based on a NAND flash-memory, which differs from the known analogues by selecting a user script with a storage device that defines the most frequently rewritten pages, journaling the most frequently rewritten pages, and merging a journal block and the corresponding data block for a read request, as well as “deferred write” blocks for rewriting operations of pages not related to the journal sible. The technical result is to extend the life of the flash memory and increase the speed of writing information to the medium by reducing the number of block erase operations, as well as increasing the speed of reading information.

State of the art

Solid-state flash storage devices are widely used in various electronic devices. Compactness, low cost, mechanical strength, low power consumption allow flash drives to remain one of the most common ways of storing and transporting digital information.

NAND flash technology has a number of features:

1) memory cells provide a limited number of write-erase cycles, after which they become unusable;

2) access for writing or reading is possible to an arbitrary number of cells, but information is erased only for a group of cells. Logical units of reading data - pages that are combined into blocks - units of erasure. Thus, data is recorded per page, and erasure is block-by-block;

3) when overwriting information in an already used memory area to ensure its correctness, this section must first be erased, that is, in case of overwriting several pages, the whole block is cleared. A large amount of resources are spent on data rewriting operations (copying unwritable block pages, erasing a block, recording new and copied data).

These features of flash-memory technology impose restrictions on the life of drives, as well as on the speed of recording information.

In order to solve these problems, the concept of logical and physical addresses is used: logical addresses are used in order to address information from the operating system, and physical addresses are used to access a physical data warehouse. Correspondence of logical and physical addresses are set in a table containing logical addresses and pointers to the corresponding physical addresses.

To reduce the number of block erase operations, and therefore, extend the life of flash drives, a solution was proposed consisting in the following: when new information is written to a busy block (overwriting), it is written to some unused block, and the busy block is cleared; then, changes are made to the correspondence table of logical and physical addresses (the logical address of the originally addressed block is assigned to the actually recorded block). With the permutation of the blocks described above, multiple access directed to the same logical block is actually performed in different physical blocks, which makes it possible to extend the service life of the flash memory, determined by the number of write operations performed in it [patent RU2243588, IPC G06F 12 / 06, G06F 12/10, published December 27, 2004].

The block from which the data was deleted is now considered an unused block, that is, it can be involved in the next write operation. An obvious improvement is the idea of taking into account the number of write-erase cycles performed for each physical block - various methods of “wear-leveling”. In [application US20070050536, IPC G06F 12/00, published March 1, 2007] the next block for recording is selected from a list sorted by the time of the last erasure and / or by the number of previously recorded recording cycles of the block. In [patent US6850443, IPC G11C 11/00, published 01.02.2005] upon detection of uneven wear, i.e. in the case when some physical blocks have more rewriting cycles than others, these blocks exchange information with the least rewritable blocks, and changes are made to the correspondence table of logical and physical addresses reflecting the exchange made.

Performance is significantly reduced and flash wear is increased in the event of multiple overwriting of a small amount of data (less than the block size). In this case, you must first copy the unwritten pages of the block, then erase the block and, finally, record both new and copied data. It was proposed to conduct page logging [patent US6938116, IPC G06F 12/16, G06F 12/00, G06F 12/02, G06F 13/42, G11C 16/02, published 08/30/2005]. All available memory is divided into 3 groups: data blocks (occupied), free blocks and journal blocks (reserved). When performing a page rewrite operation, data is not written to the requested data block, but to the journal block assigned to it. When you re-request a rewrite of this page, it is again written to the journal block (to free space). When the journal block is completely filled, a merge is performed: the latest current version of the journal page and the remaining unwritten pages of the requested block are copied to the third free block, which is assigned the logical address of the requested. The requested block is cleared. Reading data from a block containing a journaling page (s) is as follows: non-journaling pages are read at their immediate address, and reading data from journaling pages is carried out from the journal area. With this approach, the speed of the drive is also improved by reducing the number of erase operations. This solution is the closest analogue (prototype) of the proposed method of recording data to a digital information storage device based on NAND flash-memory.

Disclosure of invention

The essence of the invention as a technical solution lies in a method of recording data on a storage medium based on a flash memory of the NAND type, including

receiving a request to write data to the blank pages of a specific block and directly write data to these pages,

receiving a request to overwrite data in the filled pages of a specific block and logging data in a journal block corresponding to the requested block,

merging a data block and its corresponding journal block,

characterized in that

at the beginning of the user’s work with the drive, the user’s work scenario is automatically selected from the set of suggestions generated at the development stage during preliminary modeling of the user’s work with the drive based on the collected statistical information about the frequency of page rewriting,

journaling is performed only for the most frequently rewritten pages defined by the script selected at the beginning of the user’s session with the drive,

the merging of the data block and the corresponding journal block is performed upon receipt of a request to read a journal page from this block,

for rewriting operations of pages that are not related to journaling, a “deferred write” of blocks is performed, which consists in writing information to the buffer block instead of the addressable block until the current addressable block number or the first request to read data changes.

The invention is aimed at solving the problem of increasing the life and speed of information storage devices based on NAND flash-memory. The technical result is to extend the life of the flash memory and increase the speed of writing information to the medium by reducing the number of block erase operations, as well as increasing the speed of reading information.

At the beginning of the operation session of the information storage device, an automatic selection of a user behavior scenario is performed. Based on the selected scenario, a certain number (n) of the most frequently rewritten pages (areas) are recorded and their journaling is carried out. The merging of the latest relevant data from the journal blocks with the corresponding blocks, in contrast to the prototype, is carried out not when the journal blocks are completely filled, but when the first request for reading data is received. To rewrite the rest of the pages (not related to journaling), “delayed recording” of blocks is proposed. When a request for overwriting is received, the recording of new data is carried out in a free buffer block. If the next rewrite request refers to the same block, but to different pages, the recording is again performed in the buffer block. When you move to another block (or when you receive a read request), the pages of the source block are copied to the blank pages of the buffer block, then the source block is erased, and in the correspondence table of logical and physical addresses, the logical address of the originally addressed block is assigned to the really recorded (buffer) block.

Logging frequently used pages (small areas) allows you to reduce the number of rewriting operations (write-erase cycles) by the number of times equal to the number of pages (areas of a given size) in a block (from 64 to 512, depending on the size of the drive), which extends the service life drive, and also significantly improves the speed of recording information.

The merging of the latest relevant data from the journal block with the corresponding data block not when the journal block is full, but, unlike the analogue, when a request is made to read the data of the journal page, it can improve write speed (by eliminating unnecessary operations of the merge procedure in case of continuous series of requests for overwriting) and reading, since access to unfragmented, sequentially located data within the same block is faster.

In modern operating systems, large amounts of data are recorded in small packets in ascending order of logical addresses. The packet size is often smaller than the block size. “Delayed write” of blocks improves the service life and performance in cases of overwriting a large amount of data by reducing the number of block erase operations by a factor equal to the size of the block divided by the packet size.

Brief Description of the Drawings

FIG. 1 - Assign journal blocks to journal pages.

FIG. 2 - A generalized scheme of the data recording algorithm (logging).

FIG. 3 - An example of the state of the page history.

FIG. 4 - A generalized scheme of the algorithm for merging the journal block and the corresponding data block.

FIG. 5 - A generalized scheme of the delayed recording algorithm (“delayed recording” of blocks).

FIG. 6 - Schematic representation of a series of transactions in accordance with the "delayed record."

FIG. 7 - Schematic representation of the merger of blocks in accordance with the "delayed recording."

The implementation of the invention

User scenario selection

In the test mode of operation of the storage medium, user work with memory was modeled and statistics on the frequency of page rewriting were collected. As a result of the analysis of statistical information, various scenarios of the user’s work with the storage medium that determine the most frequently used (rewritable) memory areas are formed.

For example, these scripts may correspond to the type of file system used. Regardless of what type of file system is used, separate logical areas of small memory (1-2 pages) are most often used for recording / overwriting during operation (for example, in the FAT32 file system, these are boot recording areas BOOT, root directory ROOT, and FAT file allocation table itself). Information about in which file system the information storage device is formatted is located in the boot record area and is read at the beginning of the work session. The operation scenario corresponding to the given file system is selected.

Upon receipt of a request to write data to a specific block, it is checked whether the required pages of the block are free. Data is written directly to the free pages of the requested block. In the case of a request for rewriting (recording information in the filled pages of a block), one of two options is used: journaling (for the most frequently rewritten pages) or “delayed recording” of blocks.

Frequently Used Pages Logging

In accordance with the selected user's work scenario, a certain number (n) of the most frequently rewritable pages for journaling is determined: Ax, By, Cz ..., where A, B, C are block addresses, x, y, z are page numbers in the corresponding blocks A, B, C. Each journal page is assigned a journal (reserved) block: A ', B', C '... respectively (Fig. 1).

A generalized scheme of the data rewrite algorithm is shown in Figure 2. When a request to rewrite data is received at A.x, it is checked whether the page A.x is journaled (1). If so, it checks to see if the corresponding journal block A ’(2) is full. If block A ’is full, block A’ (3) is cleared. Then it is recorded on the first blank page of block A ’(4). If the A.x page is not a journal page, then the “delayed record” of blocks (5) works for the A.x address, Figure 5.

Merge journal block and data block

Upon receipt of a request to read data, the requested block is merged with the corresponding journal block if the journal block is nonempty. A generalized scheme of the block fusion algorithm is presented in Figure 4.

Upon receipt of a request to read data at A.x, it is checked whether the A.x page is journaled (1). If yes, then it is checked if there is data in the corresponding journal block A ’(2). If block A 'is non-empty, then the third clean block B (3) is selected, into which the last filled (actual) page of journal block A' is copied to its place (at address Bx) (4), as well as copying the data of the remaining pages of the requested block A (5); Blocks A and A ’are cleared (6); in the correspondence table of logical and physical addresses, the logical address of the initially addressed block A is assigned to the new block B and vice versa (7). After these operations, as well as if the page A.x does not belong to the journaled pages or the corresponding journal block A ’does not contain data, the data is read directly to the address A.x (8).

"Delayed recording" of blocks

To implement “deferred recording," one block is reserved, called a buffer (R).

A generalized scheme of the “delayed recording” algorithm of the blocks is shown in Figure 5. When a request is made to rewrite data at address D, writing is made to the buffer block R (1). The block number of the current transaction is stored in the variable prevD (2). If the next rewrite request relates to the same block D, but to different pages, the recording is again performed in the buffer block R (1). When changing to another block (prevD ≠ D) (3), the pages of the original block prevD (merging of blocks - 4) are copied to the empty pages of the buffer block R, then the original block prevD is cleared (5), and the logical address in the table of correspondence of logical and physical addresses The initially addressed prevD block is assigned to the actually recorded (buffer) block R (6).

The first request to read data from block D also merges the prevD and R blocks, erases the prevD, and changes the pointers in the correspondence table of logical and physical addresses.

Schematically, “deferred recording” is depicted in figures 6-7.

Claims (10)

A method of recording data to a digital information storage device based on a NAND flash memory, including receiving a request to write data to the blank pages of a specific block and directly write data to these pages, receiving a request to overwrite data in the filled pages of a specific block and logging data in a journal block corresponding to the requested block, merging a data block and its corresponding journal block, characterized in that at the beginning of the user’s work with the drive, the user’s work scenario is automatically selected from the set of proposed ones, journaling is performed only for the most frequently rewritten pages defined by the script selected at the beginning of the user’s session with the drive, the merging of the data block and the corresponding journal block is performed upon receipt of a request to read a journal page from this block, for rewriting operations of pages that are not related to journaling, a “deferred write” of blocks is performed, which consists in writing information to the buffer block instead of the addressable block until the current addressable block number or the first request to read data changes.

Images