CN114415968B - Storage system and data writing method thereof - Google Patents

Abstract

The invention discloses a storage system and a data writing method thereof, wherein the method comprises the steps of receiving a data writing request, and selecting the RAID level of data storage according to the data corresponding to the writing request and the state of the storage system; the physical disk of the storage system is divided into a plurality of storage blocks, the plurality of storage blocks from different physical disks are configured into storage groups according to the storage structures corresponding to the RAID levels, and the plurality of storage groups are configured into RAID storage groups of corresponding RAID levels; allocating a storage space according to the RAID level of data storage, determining the stripe relation of a storage group, and splitting a write request into a plurality of sub-requests according to the mapping relation between a stripe and a storage block; and executing the sub-request to finish data writing. By adopting the technical scheme, data redundancy is kept, and high-performance data writing can be provided.

Description

Storage system and data writing method thereof

technical field

The invention relates to the technical field of data storage, in particular to a storage system and a data writing method thereof.

Background technique

RAID (Redundant Array of Inexpensive Disk), that is, cheap redundant disk array, is a data redundancy backup technology commonly used in storage systems at present. According to different combinations of disks, RAID can be divided into RAID0, RAID1, RAID10, RAID3, RAID5, RAID6 and other levels.

With the development of network technology and information processing technology, the storage system contains more and more physical disks. Because RAID6 level can tolerate the damage of two disks at the same time, it can protect users to a greater extent while taking into account performance and disk space ratio. Data reliability, so it is widely used in storage devices.

The problem is that the storage system using RAID6 will have write penalties and non-consistent data updates, especially when small data writes and disk failures cause storage system degradation, the impact of RAID6 write penalties on data write performance is particularly high. obvious.

Contents of the invention

Purpose of the invention: The present invention provides a storage system and a data writing method thereof. By setting multiple RAID levels on multiple physical disks, the adaptive selection is the most suitable according to the storage space status of the storage system and the length of the written data. The RAID level is designed to maintain data redundancy while still providing high-performance data writing.

Technical solution: the present invention provides a data writing method of a storage system, comprising: receiving a data writing request, and selecting a RAID level for data storage according to the data corresponding to the writing request and the state of the storage system; The disk is divided into multiple storage blocks. According to the storage structure corresponding to the RAID level, multiple storage blocks from different physical disks are configured as storage groups, and multiple storage groups are configured as RAID storage groups of corresponding RAID levels; according to the RAID level of data storage Allocate storage space at the level, determine the stripe relationship of the storage group, split the write request into multiple sub-requests according to the mapping relationship between stripes and storage blocks; execute the sub-requests, and complete data writing.

Specifically, the RAID level configured by the storage system includes at least one of RAID0, RAID1, RAID5, and RAID6.

Specifically, the selection of the RAID level of data storage includes: if the state of the storage system is that there is no redundant storage space configured or the redundant storage space is insufficient, selecting the RAID0 level; if the state of the storage system is a single redundant storage space , the write performance determined by the data length determines two copies of RAID1 or RAID5; if the state of the storage system is a dual-redundant storage space, the write performance determined by the data length determines three copies of RAID1 or RAID6; the redundant storage space refers to Mirror copy storage space or checksum storage space.

Specifically, the execution of the sub-request further includes: recording a correspondence between a logical address in which data is written and a physical address in the storage space.

Specifically, the physical disk of the storage system is divided into multiple storage blocks with equal storage sizes.

Specifically, it also includes data migration. The steps include: reading data from the storage space, and reorganizing according to the stripe relationship; allocating the organized data to a new storage space according to the corresponding RAID level, and writing and updating the data The corresponding relationship between the logical address of data writing and the physical address of the storage space, and at the same time release the storage space occupied before the migration.

The present invention also provides a storage system, including: a RAID deployment module, a data writing module, and a disk array, wherein: the RAID deployment module is used to receive a data writing request, and according to the data corresponding to the writing request and the storage system State, select the RAID level of data storage; The data writing module is used to allocate storage space according to the RAID level of data storage, and determine the stripe relationship of the storage group, and write the request according to the stripe and storage block The mapping relationship is divided into multiple sub-requests; the sub-requests are executed to complete data writing; the disk array, the physical disk is divided into multiple storage blocks, and multiple storage blocks from different physical disks are selected according to the storage structure corresponding to the RAID level It is configured as a storage group, and multiple storage groups are configured as RAID storage groups of corresponding RAID levels.

Specifically, the RAID deployment module is used to select the RAID0 level if the state of the storage system is that there is no redundant storage space configured or the redundant storage space is insufficient; if the state of the storage system is a single redundant storage space, the data The write performance determined by the length determines two copies of RAID1 or RAID5; if the state of the storage system is dual redundant storage space, the write performance determined by the data length determines three copies of RAID1 or RAID6; the redundant storage space refers to mirror copy storage space or parity storage space.

Specifically, it includes a space management module for allocating storage space and recording the correspondence between the logical address where data is written and the physical address of the storage space.

Specifically, it includes a data migration module, which is used to read data from the storage space and reorganize according to the stripe relationship; allocate storage space for the organized data according to the corresponding RAID level, and write data; the space management module uses Based on the correspondence between the logical address where the record data is written and the physical address of the storage space, the free storage space is updated and released.

Beneficial effect: Compared with the prior art, the present invention has the following significant advantages: while maintaining data redundancy, it can still provide high-performance data writing.

Description of drawings

Figure 1a is a schematic diagram of the capitalization mode of RAID6.

Figure 1b is a schematic diagram of the RAID6 lowercase mode.

Figure 2a is a schematic diagram of the uppercase method of the degraded stripe.

Fig. 2b is a schematic diagram of a lowercase mode of degraded stripes.

FIG. 3 is a schematic structural diagram of a storage system provided by the present invention.

Fig. 4 is a schematic diagram of the data mapping relationship provided by the present invention.

Fig. 5 is a schematic diagram of the data migration method provided by the present invention.

FIG. 6 is a schematic diagram of the hardware of the storage system provided by the present invention.

Detailed ways

The technical solution of the present invention will be further described below in conjunction with the accompanying drawings.

Before describing the solutions provided by the present invention in detail, a brief description will be given of the relevant terms and principles involved in the present invention.

Stripe: RAID is composed of multiple disks. The stripe technology distributes and stores data in blocks among multiple disks, so that data can be processed concurrently. In this way, writing and reading data can be performed concurrently on multiple disks, thereby extending I/O performance well.

RAID0: It is a simple data striping technology without data redundancy, which stores data scattered in all disks, and realizes concurrent read and write of multiple disks in an independent access manner, thereby achieving multiple bandwidth performance , the read and write speed is the fastest among all RAID levels. However, RAIDO does not have redundant information such as mirroring and verification, so it cannot provide data protection functions. Once the disk is damaged, all data will be lost.

Mirrored RAID: The mirrored RAID striping technology writes data to multiple disks in a consistent manner based on the number of copies of the data. Mirrored RAID achieves data redundancy by sacrificing a certain amount of storage space. If the number of mirror copies is 2, the overall space utilization is only 50%. Although mirrored RAID has low space utilization, it provides good data protection. A damaged disk will not cause data loss, and has little impact on data read and write performance.

RAID5 and RAID6: These two RAID levels obtain the verification information by encoding the data on the same stripe according to a certain format, and then write the data to the corresponding data disk, while the verification information is written into the verification plate. RAID5 has only one parity disk and can tolerate damage to one disk in the storage pool. And RAID6 has two parity disks, which can tolerate damage to two disks at the same time. Generally, when configuring a storage pool, configure the data disks to a power of 2, such as 4+1, which means a RAID5 storage pool with 4 data disks and 1 parity disk; and 4+2, which means 4 Data disk, RAID6 storage pool with 2 check disks. The read performance of RAID5 and RAID6 has the same high bandwidth as RAID0 and RAID1 when there is no disk damage. However, when data is written, due to the need to calculate the checksum data, the write performance will be greatly affected. However, if there are faulty disks on the RAID5 and RAID6 stripes, the read and write performance of data will be greatly affected. Generally, there are two ways to calculate the verification data, uppercase and lowercase. In order to illustrate the impact of different writing methods on performance, the following is a brief introduction to uppercase, lowercase and the principle of reading and writing when there is a faulty disk with the help of the accompanying drawings.

FIG. 1a is a schematic diagram of the principle of the 4+2 stripe uppercase mode of RAID6; FIG. 1b is a schematic diagram of the principle of the 4+2 lowercase mode of RAID6. Specifically, as shown in Figure 1a and Figure 1b, when new data is written to the stripe D1, in order to calculate the new verification values P and Q of the stripe, it can be realized by using uppercase or lowercase, where the uppercase , it is necessary to pre-read the data of D2, D3, and D4, and then calculate the verification values P and Q according to RAID6, and obtain NewP1=NewD1^D2^D3^D4; NewQ1=k1*NewD1^k2*D2^k3* D3^k4*D4. After the calculation is completed, write NewD1, NewP1, and NewQ1 into the stripe to complete the data writing. In the lowercase mode, it is necessary to pre-read the data of D1, P and Q, and then calculate the new P and Q according to the equivalence equation of RAID6 uppercase, that is, NewP1=NewD1^D1^P1, New Q1=k1*NewD1^ k1*D1^Q1.

Figure 2 is a schematic diagram of the principle of RAID6 4+2 stripes with faulty disk data writing. There are two situations. Figure 2a shows that the newly written data just falls on the faulty disk. In this scenario, you can only use Uppercase method; Figure 2b shows that the newly written data falls on other normal disks. In this scenario, it can only be realized in lowercase.

According to the principle of RAID6, it can be known that there are two parity disks, so it can tolerate the damage of two disks at the same time, while taking into account performance and disk space ratio, it can also protect the reliability of user data to a greater extent. However, no matter whether the data is written in uppercase or lowercase, it is inevitable to calculate the new verification data by pre-reading the old data on the data disk or the verification disk, and write new data every time. Both need to be written to the verification disk. This additional read-ahead and write verification disk is the write penalty of RAID5 and RAID6. The higher the write penalty, the greater the impact on performance. In addition, according to the data writing method of RAID5 or RAID6, the written data will be distributed on different disks. Once a writing error occurs on a certain disk during the writing process, while other disks are successfully written, the entire stripe The data will be destroyed, and the correct data can no longer be calculated through P or Q. This is the problem of inconsistent updates.

Referring to Fig. 3, the present invention provides a data writing method of a storage system, comprising: receiving a data writing request, and selecting the RAID level of data storage according to the data corresponding to the writing request and the state of the storage system; The physical disk is divided into multiple storage blocks. According to the storage structure corresponding to the RAID level, multiple storage blocks from different physical disks are configured as storage groups, and multiple storage groups are configured as RAID storage groups of corresponding RAID levels; The RAID level allocates storage space, determines the stripe relationship of the storage group, splits the write request into multiple sub-requests according to the mapping relationship between stripes and storage blocks; executes the sub-requests, and completes data writing.

In the embodiment of the present invention, the RAID level configured by the storage system includes at least one of RAID0, RAID1, RAID5 and RAID6.

In specific implementation, some or all of the storage resources in the system can be selected to create a storage pool, and the data redundancy mode can be configured for the storage pool. For example, it can be configured as 4+2 mode, which means that the storage system has dual redundancy on the same stripe. , allowing damage to two disks at the same time.

In the embodiment of the present invention, the physical disk of the storage system is divided into multiple storage blocks with equal storage sizes.

In a specific implementation, each physical disk in the storage pool is divided according to a fixed size, and all physical disks are divided into storage blocks (Chunks) of equal size. The fixed size can be set according to actual needs and is not specifically limited. The storage blocks divided by all physical disks are virtualized into a storage pool for subsequent space management allocation.

In a specific implementation, space management modules of different RAID levels are deployed according to the redundancy mode (redundant storage space). For example, if it is configured as 4+2 redundancy mode, the following RAID levels can be deployed: RAID0; two-copy RAID1; three-copy RAID1; RAID5; RAID6. Among them, RAID0 is used for data writing when the entire storage system has no redundant disks; two copies of RAID1 and RAID5 are used for data writing when the entire storage system has only one redundant disk. More specifically, two Replica RAID1 is used for small-block data writes, while RAID6 is used for large-block data writes. During actual implementation, several or all of the RAID levels can be selected to be deployed, and the one with the highest performance is selected to complete data writing according to the actually deployed RAID levels.

In a specific implementation, according to the redundancy mode, storage blocks from different disks are formed into RAID storage groups (Chunk RAID Groups). Taking RAID6 in 4+2 redundancy mode as an example, it is necessary to select a storage block from 6 different physical disks to form a storage group (Chunk Group), so that data can fall on 6 different physical disks, ensuring that When two disks fail, data can still be read and recovered normally. Multiple storage groups make up the RAID storage group for this RAID level.

In a specific implementation, the storage groups in the RAID storage group are further divided into finer-grained small storage blocks (Blocks), and the small storage blocks from different storage blocks of the same storage group form a stripe.

In a specific implementation, one or more stripes are the basic unit for storing data in a logical storage unit (LUN). The LUN is a storage unit that can be directly mapped to the host to read and write data. When processing user read and write requests and performing data migration, the LUN applies for space from the storage system, releases space, and migrates data according to the conditions of the space management module. carried out with the unit.

Refer to Figure 4, which lists two storage systems that deploy three copies of RAID1 and RAID6. The logical storage unit LUN receives the host data write request, and selects the appropriate space management module to allocate storage space according to the length of data written and the current state of the storage system. After the data is written, record the mapping from the logical address to the physical address of the LUN. The space management module obtains storage blocks from different disks in the storage pool, forms storage groups according to the RAID level, and further divides them into smaller storage blocks of finer granularity and allocates them to LUNs. The space management module manages the free space and maintains the mapping relationship between the storage group and each physical disk storage block. The physical disk storage pool is responsible for dividing the physical disk into storage blocks, allocating and maintaining the space usage of each disk storage block.

In a specific implementation, a data write request from the host is received, and the write request includes the initial logical address LBA to be written into the LUN and the length of the data to be written. Since the minimum unit of space management allocation and release is a small storage block defined by the storage system, before allocating space, first check whether the host write request is B1ock aligned. If the logical address and data length are not aligned, then the write request needs to be The head and tail of the file are pre-read and filled to ensure the integrity of the written data.

In the embodiment of the present invention, if the state of the storage system is that there is no redundant storage space configured or the redundant storage space is insufficient, the RAID0 level is selected; if the state of the storage system is a single redundant storage space, the write rate determined by the data length The performance determines two copies of RAID1 or RAID5; if the state of the storage system is dual redundant storage space, the write performance determined by the data length determines three copies of RAID1 or RAID6; the redundant storage space refers to the mirror copy storage space or parity storage.

In the specific implementation, it is assumed that a storage pool is created and configured as a 4+2 redundancy mode of RAID6, and all RAID levels are deployed for illustration. If the system is currently in a no-redundancy state, it will directly allocate storage space from the RAID0 space management module; if the system is currently in a single-redundancy state, it will further determine whether to use two copies of RAID1 or RAID5 (4+1) according to the data length of the write request. ) The space management module allocates storage space, and the specific judgment condition is: if the length of the written data is less than 4 Blocks, then allocate storage space from two copies of RAID1, otherwise allocate from RAID5. Taking the written data length as 1 Block as an example, under the condition of ensuring the same data redundancy, according to the principles of two-copy RAID1 and RAID5, it can be known that the two-copy RAID1 needs to write 2 times, while RAID5 needs to pre-read at least 3 disks , write 3 disks, so the write amplification is 6, and the performance of allocating storage space from RAID1 is the highest. The same is true for other storage space allocation situations. According to the storage pool disk configuration (2+2, 4+2 or 8+2, etc.) and the length of the written data, the write amplification is calculated, and the write performance is selected according to the write amplification. High RAID level and allocate corresponding storage space.

In the embodiment of the present invention, the correspondence between the logical address where data is written and the physical address of the storage space is recorded.

In the specific implementation, according to the determined RAID level of the space management module and the allocated storage space, the stripe relationship is determined, and the data writing of the entire stripe is completed after completing the data writing of multiple copies or calculating the P and Q check codes. The space management module maintains the mapping relationship between the RAID storage group and the physical disk, and the write request of the host is divided into different sub-requests according to the mapping relationship between the stripe and the physical disk, and sent to each physical disk . The write address in the host write request is the logical address LBA, the storage space allocated by the space management module is the physical address PBA, and the sizes of the LBA and PBA are aligned to the block size. The block addressing method is that the upper 8 bits are used to indicate the Chunk RAID Group ID to which it belongs, and the lower 56 bits are used to indicate the offset within the RAID storage group.

In the specific implementation, after all split sub-requests are written, the space management module records the storage space allocated by this write request and manages the free space in this module, and the logical volume LUN layer records the mapping between LBA and PBA Relationship, complete this write request.

In the specific implementation, the mapping relationship between the RAID storage group and the physical disk, the usage of the physical disk space, the block allocation and usage of the space management module, and the mapping relationship between the logical volume LUN layer LBA and the PBA and other elements of the storage system Data can have an independent storage medium, and data structures without data can use databases or other custom data types.

According to the realization principle of RAID, the effective storage space of different RAID levels is very different. The effective storage space of two-copy RAID1 is only 50% of the total disk space, and the effective storage space of three-copy RAID1 is only 50% of the total disk space. 33%. For a storage system configured as 4+2 RAID6, its effective storage space is 66% of the total disk space. Therefore, in order to ensure the writing performance, the storage space of the disk can also be fully utilized.

In the embodiment of the present invention, data migration is also included, and the steps include: reading the data from the storage space, and reorganizing according to the stripe relationship; allocating the organized data to a new storage space according to the corresponding RAID level, and performing data writing Input, update the corresponding relationship between the logical address of data writing and the physical address of the storage space, and at the same time release the storage space occupied before the migration.

In a specific implementation, the data migration module can migrate user data from a RAID with a low space utilization level to a RAID with a high space utilization level. Refer to Figure 5. This diagram takes the migration of two-copy RAID1 to 4+2 RAID6 as an example. In actual use, the storage space of RAID0, three-copy RAID1 or RAID5 can be migrated to improve redundancy or increase effective storage space.

In a specific implementation, the data migration starts, and the block list to be migrated is obtained from the RAID storage group to be migrated. The B1ock to be migrated refers to the storage space containing valid user data. Read out the user data on these effective storage spaces, and reorganize the user data according to the 4+2 stripe relationship of RAID6. Allocate a full stripe of storage space from the space management module corresponding to RAID6, calculate the P and Q check codes of the stripe with the reorganized user data, and complete the data writing of a whole stripe. After the writing is completed, the data mapping relationship of the logical volume LUN needs to be updated first, and then the free space of the RAID6 space management module is updated and the storage space occupied by the two copies of RAID1 is released.

The present invention also provides a storage system, including: a RAID deployment module, a data writing module and a disk array, wherein:

The RAID deployment module is configured to receive a data write request, and select the RAID level of data storage according to the data corresponding to the write request and the state of the storage system;

The data writing module is used to allocate storage space according to the RAID level of data storage, and determine the stripe relationship of the storage group, and split the write request into multiple sub-requests according to the mapping relationship between the stripe and the storage block; Execute the sub-request and complete the data writing;

In the disk array, the physical disk is divided into multiple storage blocks, and according to the storage structure corresponding to the RAID level, multiple storage blocks from different physical disks are configured as storage groups, and multiple storage groups are configured as RAID storage groups of corresponding RAID levels .

In the embodiment of the present invention, a space management module is included for allocating storage space and recording the correspondence between the logical address of data writing and the physical address of the storage space.

In the embodiment of the present invention, a data migration module is included, which is used to read the data from the storage space and reorganize according to the stripe relationship; allocate the organized data to the storage space according to the corresponding RAID level, and write the data; The management module is used to record the corresponding relationship between the logical address of data writing and the physical address of the storage space, update and release the free storage space.

In the embodiment of the present invention, the RAID deployment module is used to select the RAID0 level if the state of the storage system is that no redundant storage space is configured or the redundant storage space is insufficient; if the state of the storage system is single redundant storage space , the write performance determined by the data length determines RAID1 or RAID5; if the state of the storage system is a dual-redundant storage space, select RAID6; the redundant storage space refers to the mirror copy storage space or the verification storage space.

In the embodiment of the present invention, the RAID deployment module is used when the data length is less than 4 small storage blocks, select RAID1; when the data length is greater than or equal to 4 small storage blocks, select RAID5; the small storage block is supported by the storage system The data block with the smallest granularity is obtained by further dividing the storage group.

Referring to FIG. 6, the hardware of the storage system includes a storage system controller and a disk array, wherein the storage system controller is a dual-controller architecture for mutual backup and load sharing. A disk array is a disk cabinet containing several physical disks, and is connected to the storage system controller through the system bus. The storage system controller 10 and the storage system controller 20 have the same structure and function, and the two controllers are mutually backed up and load shared, which can effectively ensure the safety and reliability of the storage system and ensure the write operation performance of the storage system .

The disk array 30 includes several physical disks to provide actual physical storage space for the storage system. Various storage media can be used, and even different storage media can be mixed and matched. The storage system creates a storage pool based on the disk array and deploys a corresponding storage system controller.

Claims (8)

1. A data writing method of a storage system, comprising:

receiving a data writing request, and selecting a RAID level of data storage according to data corresponding to the writing request and the state of a storage system; the physical disk of the storage system is divided into a plurality of storage blocks, the plurality of storage blocks from different physical disks are configured into storage groups according to the storage structures corresponding to the RAID levels, and the plurality of storage groups are configured into RAID storage groups of corresponding RAID levels; if the state of the storage system is that redundant storage space is not configured or insufficient, selecting RAID0 level; if the state of the storage system is a single redundant storage space, determining two copies of RAID1 or RAID5 according to the write-in performance determined by the data length; if the state of the storage system is a dual-redundancy storage space, determining three-copy RAID1 or RAID6 according to the write performance determined by the data length; the redundant storage space refers to a mirror image copy storage space or a check storage space;

allocating a storage space according to the RAID level of data storage, determining the stripe relation of a storage group, and splitting a write request into a plurality of sub-requests according to the mapping relation between a stripe and a storage block;

and executing the sub-request to complete data writing.

2. The data writing method of the storage system according to claim 1, wherein the RAID level of the storage system configuration includes at least one of RAID0, RAID1, RAID5, and RAID6.

3. The data writing method of the storage system according to claim 2, wherein the executing the sub-request further comprises:

and recording the corresponding relation between the logical address written by the data and the physical address of the storage space.

4. The data writing method of the storage system according to claim 3, wherein the physical disk of the storage system is divided into a plurality of storage blocks with equal storage size.

5. The data writing method of the storage system according to claim 4, further comprising data migration, the steps comprising:

reading data from the storage space, and reorganizing the data according to a stripe relation;

distributing new storage space for the organized data according to corresponding RAID level, writing the data, updating the corresponding relation between the logic address written in the data and the physical address of the storage space, and releasing the occupied storage space before migration.

6. A storage system, comprising: RAID deploys module, data write-in module and disk array, wherein:

the RAID deployment module is used for receiving a data write-in request and selecting the RAID level of data storage according to the data corresponding to the write-in request and the state of the storage system; if the state of the storage system is that redundant storage space is not configured or insufficient, selecting RAID0 level; if the state of the storage system is a single redundant storage space, determining two copies of RAID1 or RAID5 according to the write-in performance determined by the data length; if the state of the storage system is a double-redundancy storage space, determining three-copy RAID1 or RAID6 according to the write-in performance determined by the data length; the redundant storage space refers to a mirror image copy storage space or a check storage space;

the data writing module is used for allocating storage space according to the RAID level of data storage, determining the stripe relation of a storage group, and splitting a writing request into a plurality of sub-requests according to the mapping relation between a stripe and a storage block; executing the sub-request to finish data writing;

in the disk array, a physical disk is divided into a plurality of storage blocks, the plurality of storage blocks from different physical disks are configured into storage groups according to a storage structure corresponding to the RAID level, and the plurality of storage groups are configured into RAID storage groups of corresponding RAID levels.

7. The storage system according to claim 6, comprising a space management module for allocating the storage space and recording the correspondence between the logical address of the data write and the physical address of the storage space.

8. The storage system according to claim 7, comprising a data migration module for reading data from the storage space and reorganizing the data according to a stripe relationship; distributing storage space for the organized data according to corresponding RAID levels, and writing the data; the space management module is used for recording the corresponding relation between the logical address written in the data and the physical address of the storage space, and updating and releasing the free storage space.

Images