CN102521068B - Reconstructing method of solid-state disk array - Google Patents

Abstract

The invention discloses a reconstructing method of a solid-state disk array, and the reconstructing method comprises the following steps: (1), constructing a storage system; (2), when a member solid-state disk in the system generates a fault, taking out the fault disk, inserting a new replacing disk, and starting the solid-stage disk array to begin carrying out reconstruction; and (3), writing back data. In addition, the reconstructing method further comprises a data reading-writing process at a reconstruction stage and a data reading-writing process at a writing-back stage. The reconstructing method can be used for the solid-state disk arrays which belong to the grades of RAID1 (redundant array of inexpensive disks 1), RAID10, RAID5, RAID6 and the like and have redundant data, and is applicable to the construction of a high-performance, high-capacity and high-reliability large storage system.

Description

A Reconstruction Method of Solid State Disk Array

technical field

The invention belongs to the field of computer data storage, and in particular relates to a reconstruction method of a solid-state disk array.

Background technique

Solid State Drive (SSD) is a device that uses semiconductor memory chips to store data. It does not have the rotating and moving mechanical parts of traditional hard disks for reading and writing. It has fast access speed, low energy consumption, high random access and Good shock resistance and other characteristics. Today, when the performance of traditional disk storage systems is stagnant, solid-state disks are like a wonderful flower blooming in more and more commercial applications, solving the I/O bottleneck and energy consumption problems of high-performance data centers, and are the current and future storage solutions. The hot spot and core of field research.

In recent years, MLC (Multi-Level Cell) technology has improved the capacity of solid-state drives and greatly reduced its cost. However, since each MLC storage unit stores more bits, the structure is relatively complicated, and the probability of error is very high. Large, performance and reliability are lower than SLC (Single-Level Cell, single-layer unit). In order to improve the performance of SSDs, we can use parallel I/O architectures, such as multi-channel and interleaved access technology, to increase bandwidth by allowing concurrent I/O operations of multiple memory chips. However, reliability has always been a major design issue. A key issue in large-scale storage systems. Current NAND flash products use error-checking codes to ensure reliability, such as Hamming Code, Reed-Solomon Code, and BCH code (abbreviations for Bose, Chaudhuri, and Hocquenghem), etc. , but these error-checking codes require high hardware complexity, and also cause read and write delays. Another way to improve reliability is storage-level redundancy. The standard block interface of solid-state disks can use existing redundancy mechanisms and interleaved access solutions without any modification-disk array RAID (Redundant Array of Independent Disks) , among which the RAID-5 level provides better performance and reliability, and is widely used in the commercial field.

When a member SSD in the SSD array fails and a replacement disk is added, the SSD array can automatically switch to the reconstruction mode, and the reconstruction process continuously rebuilds the data blocks on the faulty disk to the new one without interrupting the application service. replacement plate. The erasing and erasing cycle limit of SSDs and the write load balancing characteristics of SSD arrays will lead to correlation failures. Specifically, multiple devices reach the state of writing and erasing limits at the same time, showing an unrecoverable high bit error rate. Of particular concern is the failure correlation when the SSD array is in the rebuilding state. The high bit error rate caused by the write request on the surviving disk will cause a large amount of data loss. In addition, the existing Linux operating system SSD array control software is the same as the traditional disk array control software, and the data reconstruction method does not take into account the random access characteristics of the SSD, resulting in repeated reading of the same data block during the reconstruction process and read-write process. Not only is the bandwidth of the array wasted, but also the reconstruction time is long, which affects the read and write performance and reliability of the storage system.

Contents of the invention

The invention proposes a data reconstruction method of a solid-state disk array, which utilizes the very good random performance of the solid-state disk to allow the reconstruction position to follow the read load movement, give priority to rebuilding the stripe at the read request position, and reduce the read requests on the surviving solid-state disk during the reconstruction process The number of times, taking into account the characteristics of the correlation failure caused by RAID load balancing and the limit of the number of writes and erases, redirects the write load to prevent the high BER (Bit Error Rate, bit error rate) from recurring during the rebuilding process of the SSD array. For data loss caused by failure, write reconstruction can also reduce the impact of load on reconstruction, which solves the problems of long reconstruction time and low reliability of storage system in the reconstruction process of existing solid-state disk arrays.

The specific technical scheme adopted to realize the object of the present invention is as follows:

A rebuilding method of a solid state disk array of the present invention is characterized in that it comprises the following steps:

(1) Construction of storage system

First, add an additional disk outside the SSD array, which is used to store redirected write data during the reconstruction process;

Secondly, set the bitmap log table and the write redirection metadata Hash table in the memory. The bitmap log table is used to record the stripes that are reconstructed preferentially following the read load, and the metadata Hash table is used to record the key information of redirected write requests. Including the starting logical address of the stripe where the request is located, the number of the solid-state disk to be written to, the verification disk number of the stripe, and the logical block address redirected to the additional disk;

(2) The reconstruction process of the SSD array

When a member SSD in the system fails, take out the faulty disk, insert a new replacement disk, and start the SSD array to start rebuilding. The specific process is as follows:

(2.1) Initialize the bitmap log table, each bit of the log table corresponds to a strip of the solid-state disk array, each bit of the bitmap log table is initialized to 0, and continue step (2.2);

(2.2) the reconstruction process judges whether the corresponding bit of the bitmap log table is 1 according to the initial logical address of the strip that needs to be reconstructed at present, if it is, then continue step (2.3), otherwise proceed to step (2.4);

(2.3) The stripe has been rebuilt, skip the rebuilding of the stripe, obtain the starting logical address of the next rebuilt stripe, and go to step (2.5);

(2.4) The stripe has not been rebuilt. Get a free stripe, read the data at the location of the stripe on all surviving disks, perform an XOR operation on these data, and write the calculation result to the corresponding SSD In the position, the corresponding bit in the bitmap log table is 1, and the step (2.5) is continued;

(2.5) Repeat the above steps (2.2)--(2.4) until all the stripes are rebuilt, releasing the resources occupied by the bitmap log table;

(3) Data write back

(3.1) Judging whether the write redirection metadata Hash table is empty, if not empty then continue to step (3.2), otherwise proceed to step (3.4);

(3.2) Take out the M records with the same stripe starting logical address but different logical disk numbers from the write redirection metadata Hash table, read the data corresponding to these M records from the additional disk, and write them back In the rebuilt SSD array, the writing process is consistent with the normal read and write process of the SSD array, and the check block needs to be updated, and continue to step (3.3);

(3.3) delete these records from the Hash table, and proceed to step (3.1);

(3.4) The write-back process of all redirected data is completed, and the resources occupied by the Hash table are released.

A rebuilding method of a solid-state disk array of the present invention is characterized in that the specific process of reading and writing data in the rebuilding stage is:

(4.1) Judge the state of the array, if it is in the reconstruction state then continue to step (4.2), otherwise go to step (4.7);

(4.2) Judge the type of the read-write request issued by the upper layer, if it is a read request, then continue to step (4.3), otherwise go to step (4.8);

(4.3) Process the user's read request, obtain the stripe S corresponding to the read request according to the RAID data organization method, and query the write redirection according to the starting logical address C of the stripe S and the logical disk number N of the request in the stripe S Metadata Hash table, if hit then continue to step (4.4), otherwise go to step (4.5);

(4.4) The latest data to be read is in the additional disk, according to the logical block address on the additional disk in the hit metadata entry, the latest data in the additional disk is read and returned to the user, and the end is completed;

(4.5) The data to be read is in rebuilding the solid-state disk array, query the corresponding bit of the bitmap log table corresponding to the initial logical address C, if it is 0, then proceed to step (4.6), otherwise proceed to step (4.7);

(4.6) The strip S where the read request is located has not been rebuilt, read all the data blocks located in the strip S on the surviving disk from the underlying solid-state disk, and record them as D(i), where i=1, 2, 3...n-1, n is the number of member solid-state disks in the array, use XOR to calculate the data D(n) on the replacement disk, that is Return the data block D={D(i)|i=1, 2, 3...n} that needs to be read by the user's read request to the user, and write D(n) into the replacement solid-state disk at the same time, set The corresponding bit in the graph log table is 1, end;

(4.7) Process according to the normal data reading flow, end;

(4.8) Process the user's write request. For the data to be written D', obtain the stripe S' corresponding to the write request according to the RAID data organization method, calculate the starting logical address C' of the stripe S', and write the request in the stripe Logical disk number N' in S', verification disk number P' of the stripe, if the stripe is full and written, then enter step (4.9), otherwise proceed to step (4.10);

(4.9) The write data block has covered the whole stripe buffer, and the data D' to be written is filled in the stripe buffer C' (N), and then the step (4.12) is changed over to;

(4.10) Write the data block and do not cover the whole stripe buffer, check the corresponding bit of the bitmap log table according to the stripe starting logical address, if it is 0, then continue the step (4.11), otherwise go to the step (4.12);

(4.11) The stripe has not been rebuilt, read all the data blocks located in the stripe S' on the surviving disk from the solid-state disk array, and record them as D'(i), where i=1, 2, 3 ...n-1, n is the number of member solid-state disks in the array, use the XOR operation XOR to calculate the data D'(n) on the replacement disk, that is Fill the write data D' into the stripe buffer C'(N), and write D'(n) to the replacement SSD at the same time, set the corresponding bit in the bitmap log table to 1, and go to step (4.12) ;

(4.12) Redirect the data in the stripe buffer C'(N) to the additional disk. The additional disk adopts the method of sequential appending, and searches for write redirection according to the stripe start logical address C' and logical disk number N' Metadata Hash table, if hit, transfer to (4.13), otherwise transfer to (4.14);

(4.13) Update the metadata table of the hit, revise the logical block address of the additional disk in the hit table entry, mark the old data block corresponding to the data block in the additional disk as invalid, and end;

(4.14) Add the metadata entry of the redirected write request to the Hash table, and end.

A rebuilding method of a solid-state disk array of the present invention is characterized in that the specific process of reading and writing data in the write-back stage is:

(5.1) Wait for the user request from the upper layer to determine whether the array is in the write-back stage, if not, proceed according to the normal data read and write process, and if so, continue to step (5.2);

(5.2) Judge the type of request, if it is a read request, then proceed to step (5.3), otherwise proceed to step (5.6);

(5.3) Find the write redirection metadata Hash table, if hit then go to step (5.4), otherwise go to step (5.5);

(5.4) Send the read request to the additional disk, read data from the additional disk and return to the user, and turn to step (5.1);

(5.5) send the read request to the solid-state disk array, read data from the solid-state disk array and return to the user, and turn to step (5.1);

(5.6) Write the data into the solid-state disk array that has been rebuilt according to the normal data processing flow of the RAID, look up the write redirection metadata Hash table, if hit then proceed to step (5.7), otherwise proceed to step (5.1);

(5.7) Delete the hit metadata entry from the Hash table, and go to step (5.1).

A rebuilding method of a solid-state disk array according to the present invention is characterized in that the write request on the additional disk adopts a sequential append writing method, and when a redirected write request is updated, new data is sequentially appended to In the additional disk, modify the corresponding metadata to mark the old data in the additional disk as invalid.

A reconstruction method of a solid-state disk array of the present invention is characterized in that the redirection metadata Hash table is composed of four fields, and each field is sequentially the initial logical address S_LBA of the stripe, the written solid-state disk number DiskNo Each entry in the Hash table is used to record the key information of a redirected write request.

The present invention is based on a ubiquitous phenomenon that the existing Linux operating system solid state disk array control software is the same as the traditional disk array control software, and the data reconstruction method does not take into account the random access characteristics of the solid state disk, resulting in too long reconstruction time. The erase-write cycle limits of SSDs and the write load-balancing characteristics of SSD arrays will lead to coherent failures, resulting in poor reliability during SSD rebuilds. The present invention proposes a data reconstruction method for a solid-state disk array, allowing the reconstruction position to follow the read load movement, prioritizing the reconstruction of the stripe at the read request position, reducing the number of read requests on the surviving solid-state disk during the reconstruction process, and redirecting the write load at the same time. On the one hand, it prevents the data loss caused by the failure of the surviving SSD due to high BER during the reconstruction process, and on the other hand, it reduces the impact of user load on the reconstruction process. The present invention mainly includes five specific steps of construction of the storage system, data reading and writing in the rebuilding stage, rebuilding of the solid-state disk array, data writing back, and data reading and writing in the writing-back stage, which solves the problem of long rebuilding time of the existing solid-state disk array. , The problem of low reliability of the storage system during the reconstruction process is suitable for building a large storage system with high performance, high reliability and high capacity.

Description of drawings

Fig. 1 is a schematic diagram of the software and hardware architecture of the storage system of the present invention;

Fig. 2 is the structural representation of writing redirection metadata Hash table of the present invention;

Fig. 3 is a schematic diagram of the reconstruction process of the present invention;

Fig. 4 is a schematic diagram of data writing back in the present invention;

FIG. 5 is a schematic diagram of a data reading and writing process flow in the present invention.

Detailed ways

The present invention will be described in further detail below in conjunction with the accompanying drawings.

Fig. 1 is a schematic diagram of the software and hardware system architecture of the present invention: an additional hard disk needs to be added to the solid-state disk array for hardware deployment; the software structure mainly includes a reconstruction module, a write-back module and a data processing module, and these modules rely on bitmap logs Table and redirection metadata Hash table.

A rebuilding method of a solid-state disk array of the present invention is composed of a rebuilding stage and a data write-back stage in sequence, mainly including the construction of a storage system, data reading and writing in the rebuilding stage, rebuilding of a solid-state disk array, data write-back, and a write-back stage The five specific steps of data reading and writing.

(1) Construction of storage system

First, an additional disk is added outside the original SSD array, and the additional disk is used to store redirected write data during the reconstruction process.

Secondly, set the bitmap log table and the write redirection metadata Hash table in the memory. The bitmap log table is used to record the stripes that are reconstructed preferentially following the read load, and the metadata Hash table is used to record the key information of redirection write requests. As shown in Figure 2, the redirection metadata Hash table consists of four fields, each of which is the start logical address S_LBA of the stripe, the disk No. of the solid-state disk to be written, the parityNo. Each entry in the Hash table is used to record the key information of a redirected write request.

In addition, the write request on the additional disk adopts the write method of sequential appending. When a redirected write request is updated, the new data is sequentially appended to the additional disk, and the old data in the additional disk will be appended by modifying the corresponding metadata. Identified as invalid.

(2) The reconstruction process of the SSD array

Figure 3 is a schematic diagram of the reconstruction steps of the solid state disk array, and the following process is carried out: when the reconstruction process rebuilds a certain stripe, it first judges whether the corresponding bit in the bitmap log table is 1, and if so, skips the reconstruction of the stripe , otherwise read data from all surviving disks to rebuild the stripe. Repeat this process until all the stripes are rebuilt, and the rebuilding ends. The specific steps are as follows:

(2.1) Initialize the bitmap log table, each bit of the log table corresponds to a strip of the solid-state disk array, each bit of the bitmap log table is initialized to 0, and continue step (2.2);

(2.2) the reconstruction process judges whether the corresponding bit of the bitmap log table is 1 according to the initial logical address of the strip that needs to be reconstructed at present, if it is, then continue step (2.3), otherwise proceed to step (2.4);

(2.3) The stripe has been rebuilt, skip the rebuilding of the stripe, obtain the starting logical address of the next rebuilt stripe, and go to step (2.5);

(2.4) The stripe has not been rebuilt. Get a free stripe, read the data at the location of the stripe on all surviving disks, perform an XOR operation on these data, and write the calculation result to the corresponding SSD In the position, the corresponding bit in the bitmap log table is 1, and the step (2.5) is continued;

(2.5) Repeat the above steps (2.2)-(2.4) until all the stripes are rebuilt, releasing the resources occupied by the bitmap log table.

(3) Data reading and writing in the reconstruction phase

Figure 5 shows the flow chart of data processing during rebuilding and data write-back. When the SSD array is in the rebuilding state, the read-write processing module waits for the request from the upper layer user. If it is a read request, it first judges whether the request hits the Hash table , if it hits, you need to read the latest data from the additional disk and return it to the user, otherwise perform the following operations: determine whether the bit corresponding to the stripe where the request is located in the bitmap log table is 1, if it is 1, it means that the stripe has been After rebuilding, read the required data from the SSD array and return it to the user, otherwise read all the data on the surviving disk to rebuild the stripe, set the corresponding bit in the bitmap log table to 1, and return the data read by the user . If it is a write request, determine whether the stripe is full for writing, if it is full for writing, fill the write data into the stripe buffer, append the data in the buffer to the additional disk in order, and update the redirection element Data Hash table, otherwise perform the following operations: determine whether the bit corresponding to the stripe where the request is located in the bitmap log table is 1, if it is 1, it means that the stripe has been rebuilt, and read the location of the current write request from the SSD array The old data at the location is put into the stripe buffer, otherwise, read out all the data on the surviving disk to rebuild the stripe, set the corresponding bit in the bitmap log table to 1, and then fill the new data into the stripe buffer, and The data in the buffer is sequentially appended to the additional disk, and the redirection metadata Hash table is updated. The specific process is:

(3.1) Judge the state of the array, if it is in the reconstruction state, then continue to step (3.2), if it is in the data write-back state, then go to (5.1), otherwise go to step (3.7);

(3.2) Judge the type of the read-write request issued by the upper layer, if it is a read request, then continue to step (3.3), otherwise proceed to step (3.8);

(3.3) Process the user's read request, and obtain the stripe S corresponding to the read request according to the RAID data organization method. Assume that the starting logical address of the stripe is C, and the logical disk number of the request in the stripe is N. According to the stripe Start logical address C and logical disk number N query and write redirection metadata Hash table, if hit, continue to step (3.4), otherwise go to step (3.5);

(3.4) The latest data to be read is in the additional disk, according to the logical block address on the additional disk in the hit metadata entry, the latest data in the additional disk is read and returned to the user, and the end is completed;

(3.5) The data to be read is in rebuilding the solid-state disk array, query the corresponding bit of the bitmap log table corresponding to the initial logical address C, if it is 0, then proceed to step (3.6), otherwise proceed to step (3.7);

(3.6) The strip S where the read request is located has not been rebuilt, read all the data blocks located in the strip S on the surviving disk from the underlying solid-state disk, and record them as D(i), where i=1, 2, 3...n-1, n is the number of member solid-state disks in the array, use XOR to calculate the data D(n) on the replacement disk, that is Assume that the data block to be read by the user's read request is D={D(i)|i=1, 2, 3...n}, return D to the user, and write D(n) to the replacement solid state disk at the same time , set the corresponding bit in the bitmap log table to 1, and end;

(3.7) Process according to the normal data reading flow, end;

(3.8) Process the user's write request, assuming that the data to be written is D', obtain the stripe S' corresponding to the write request according to the RAID data organization method, calculate the starting logical address of the stripe S' as C', write the request The logical disk number in the stripe S' is N', and the verification disk number of the stripe is P'. If the stripe is full and written (OVERWRITE), then enter step (3.9), otherwise proceed to step (3.10 );

(3.9) The write data block has covered the entire stripe buffer, and the write data D' is filled in the stripe buffer C' (N), and then proceeds to step (3.12);

(3.10) The write data block does not cover the entire stripe buffer, check the corresponding bit of the bitmap log table according to the stripe S' initial logical address C ', if it is 0, then continue to step (3.11), otherwise go to step (3.12 );

(3.11) The stripe has not been rebuilt, read all the data blocks located in the stripe S' on the surviving disk from the solid-state disk array, and record them as D'(i), where i=1, 2, 3 ...n-1, n is the number of member solid-state disks in the array, use the XOR operation XOR to calculate the data D'(n) on the replacement disk, that is Fill the write data D' into the stripe buffer C'(N), and write D'(n) into the replacement SSD at the same time, set the corresponding bit in the bitmap log table to 1, and go to step (3.12) ;

(3.12) Redirect the data in the stripe buffer C'(N) to the additional disk. The additional disk adopts the method of sequential appending, and searches for the write data according to the starting logical address C' and the logical disk number N' of the stripe S'. Redirect the metadata Hash table, if hit, transfer to (3.13), otherwise transfer to (3.14);

(3.13) update the metadata table of the hit, revise the logical block address of the additional disk in the hit table entry, mark the old data block corresponding to the data block in the additional disk as invalid, and end;

Add the metadata entry of the redirected write request to the Hash table, and end.

(4) Data write back

Figure 4 is a schematic diagram of the steps of reading out the data in the additional disk and writing it back to the solid-state disk array after the reconstruction is completed, and the following process is carried out:

The data write-back module judges whether the redirection metadata Hash table is empty, and if it is not empty, takes out multiple records with the same starting logical address but different logical disk numbers, and reads the duplicated records from the additional disk according to the key information of the records. Directed write data, write them back to the SSD array, and delete these records from the Hash table. Repeat this process until all the data is reclaimed, so as to maintain the data consistency of the solid state disk array. The specific process is:

(4.1) The reconstruction process is completed, start the data write-back process, and enter step (4.2);

(4.2) Judging whether the write redirection metadata Hash table is empty, if not empty then continue to step (4.3), otherwise proceed to step (4.5);

(4.3) Take out M records with the same stripe start logical address but different logical disk numbers from the write redirection metadata Hash table, read the data corresponding to these M records from the additional disk, and write them back In the rebuilt SSD array, the writing process is consistent with the normal read and write process of the SSD array, and the check block needs to be updated, and continue to step (4.4);

(4.4) delete these records from the Hash table, and proceed to step (4.2);

(4.5) The write-back process of all redirected data is completed, and the resources occupied by the Hash table are released.

(5) Data reading and writing in the write-back phase

Figure 5 shows the flow chart of data processing during reconstruction and data write-back. When the SSD array is in the data write-back state, the read-write processing module waits for the request from the upper layer user. If it is a read request, it judges whether the request hits the Hash Table, if hit, read data from the attached disk, otherwise read data from the SSD array. If it is a write request, first judge whether the request hits the Hash table, if it hits, delete the hit record from the Hash table, and write the data to the rebuilt SSD array, otherwise directly write the data to the rebuilt SSD on the array. The specific process is:

(5.1) Wait for the user request from the upper layer to determine whether the array is in the write-back stage, if not, proceed according to the normal data read and write process, and if so, continue to step (5.2);

(5.2) Judge the type of request, if it is a read request, then proceed to step (5.3), otherwise proceed to step (5.6);

(5.3) Find the write redirection metadata Hash table, if hit then go to step (5.4), otherwise go to step (5.5);

(5.4) Send the read request to the additional disk, read data from the additional disk and return to the user, and turn to step (5.1);

(5.5) send the read request to the solid-state disk array, read data from the solid-state disk array and return to the user, and turn to step (5.1);

(5.6) Write the data into the solid-state disk array that has been rebuilt according to the normal data processing flow of the RAID, look up the write redirection metadata Hash table, if hit then proceed to step (5.7), otherwise proceed to step (5.1);

(5.7) Delete the hit metadata entry from the Hash table, and go to step (5.1).

Claims (5)

1. a method for reconstructing for solid-state disk array, is characterized in that, comprises the steps:

(1) build storage system

First, add an additional disk outside solid-state disk array, this additional disk is used for storing the data of writing that are redirected in process of reconstruction;

Secondly, bitmap log sheet is set in internal memory and writes and be redirected metadata Hash table, bitmap log sheet is used for recording following reads the preferential band of rebuilding of load, metadata Hash table is used for recording the key message that is redirected write request, comprises the verification reel number of the initial logical address of request place band, solid-state reel number that request writes, band and is redirected to the LBA (Logical Block Addressing) on additional disk;

(2) process of rebuilding of solid-state disk array

In the time having member's solid-state disk to break down in system, take out faulty disk, insert new replacement dish, start solid-state disk array and start to rebuild, detailed process is as follows:

(2.1) initialization bitmap log sheet, a band of each corresponding solid-state disk array of log sheet, is all initialized as 0 by each of bitmap log sheet, continues step (2.2);

(2.2) process of rebuilding judges according to the current initial logical address that needs the band of rebuilding whether the corresponding position of bitmap log sheet is 1, if it is continues step (2.3), otherwise proceeds to step (2.4);

(2.3) rebuilt mistake of this band, the reconstruction of skipping this band, obtains next initial logical address of rebuilding band, proceeds to step (2.5);

(2.4) this band was not rebuild, obtain an idle band, read the data at this pillar location place on all survival dishes, these data are carried out to XOR XOR computing, the result of calculating is written to and is replaced on the position that solid-state disk is corresponding, in set figure log sheet, corresponding position is 1, continues step (2.5);

(2.5) repeat above-mentioned steps (2.2)--(2.4), until that all bands are all rebuild is complete, discharge the resource that bitmap log sheet takies;

(3) data write back

(3.1) judge that whether write redirected metadata Hash table is empty, if be not sky, continue step (3.2), otherwise proceeds to step (3.4);

(3.2) metadata Hash table, take out the identical but record that logic reel number is different of band initial logical address from writing to be redirected, from additional disk, read these and record corresponding data, and they are write back in the solid-state disk array of having rebuild, upgrade check block, continue step (3.3);

(3.3) described these records are deleted from Hash table, proceeded to step (3.1);

(3.4) completed the process that writes back of all redirected data, discharged Hash and show the resource taking.

2. the method for reconstructing of a kind of solid-state disk array according to claim 1, is characterized in that, carries out the process of reading and writing data in phase of regeneration, and the process of carrying out data read and write in step (2.4) is:

(4.1) judge the state of array, if in reconstruction state, continue step (4.2);

(4.2) judge the type of the read-write requests that upper strata issues, if read request continues step (4.3), otherwise proceed to step (4.8);

(4.3) process user's read request, obtain strips S corresponding to read request according to the Method of Data Organization of RAID, according to the initial logical address C of strips S and this request, the logic reel number N inquiry in strips S is write and is redirected metadata Hash table, if hit, continue step (4.4), otherwise proceed to step (4.5);

(4.4) latest data that will read is in additional disk, and according to the LBA (Logical Block Addressing) on additional disk in the metadata list item hitting, the latest data reading in additional disk returns to user, finishes;

(4.5) data that will read, rebuilding in solid-state disk array, are inquired about the corresponding position of bitmap log sheet that initial logical address C is corresponding, and 0 proceeds to step (4.6) if, otherwise proceeds to step (4.7);

(4.6) strips S at read request place does not have rebuilt, the data block that is arranged in strips S on all survival dishes is read out from bottom solid-state disk, be designated as respectively D (i), wherein i=1, 2, 3 ... n-1, n is the number of member's solid-state disk in array, utilize XOR XOR that the data D on replacement dish (n) is calculated, be D (n)=D (1)+zero D (2)+zero ... + zero D (n-1), the data block D={D (i) that user's read request need to be read | i=1, 2, 3 ... n} returns to user, D (n) is written to simultaneously and replaces in solid-state disk, in set figure log sheet, corresponding position is 1, finish,

(4.7) read flow process according to normal data and process, finish;

(4.8) process user's write request, for data D ' to be written, obtain strips S corresponding to write request according to the Method of Data Organization of RAID ', calculate strips S ' initial logical address C ', write request is in strips S ' in logic reel number N ', the verification reel number P ' of band, writes if this band is filled band, enter step (4.9), otherwise proceed to step (4.10);

(4.9) write data block and covered whole band buffer zone, data D ' to be written is filled in band buffer zone C ' (N '), proceed to step (4.12);

(4.10) write data block and do not cover whole band buffer zone, check the corresponding position of bitmap log sheet according to band initial logical address, 0 is continued step (4.11) if, otherwise proceeds to step (4.12);

(4.11) this band does not also have rebuilt, to on all survival dishes, be arranged in strips S ' data block read out from solid-state disk array, be designated as D ' (i) respectively, wherein i=1, 2, 3 ... n-1, n is the number of member's solid-state disk in array, utilize XOR XOR that the data D ' on replacement dish (n) is calculated, be D ' (n)=D ' (1)+zero D ' (2)+zero ... + zero D ' (n-1), to write data D ' and be filled into band buffer zone C ' (N '), D ' (n) is written to and is replaced in solid-state disk simultaneously, in set figure log sheet, corresponding position is 1, proceed to step (4.12),

(4.12) by the data redirection in band buffer zone C ' (N ') in additional disk, the mode that additional disk employing order is appended, according to strips S ' initial logical address C ' and logic reel number N ' search to write and be redirected metadata Hash table, if hit, proceed to (4.13), otherwise proceed to (4.14);

(4.13) upgrade the metadata table hitting, the LBA (Logical Block Addressing) that additional disk in list item is hit in amendment, it is invalid that old data block corresponding this data block in additional disk is designated, and finishes;

(4.14) the metadata list item of this redirected write request is added in Hash table, finish.

3. the method for reconstructing of a kind of solid-state disk array according to claim 1 and 2, is characterized in that, described in write back the detailed process of the reading and writing data in stage, the process of carrying out data read and write in step (3.2) is:

(5.1) wait for that upper strata issues user's request, judges that array, whether in writing back the stage, if not processing by normal reading and writing data flow process, if it is continues step (5.2);

(5.2) type of judgement request, if read request proceeds to step (5.3), otherwise proceeds to step (5.6);

(5.3) search and write redirected metadata Hash table, if hit, proceed to step (5.4), otherwise proceed to step (5.5);

(5.4) this read request is sent to additional disk, from additional disk, reading out data returns to user, goes to step (5.1);

(5.5) this read request is sent in solid-state disk array, from solid-state disk array, reading out data returns to user, proceeds to step (5.1);

(5.6) by the normal flow chart of data processing of RAID, data are write in the solid-state disk array of having rebuild, searched to write and be redirected metadata Hash table, if hit, proceed to step (5.7), otherwise proceed to step (5.1);

(5.7) the metadata list item hitting is deleted from Hash table, proceeded to step (5.1).

4. the method for reconstructing of a kind of solid-state disk array according to claim 1 and 2, it is characterized in that, the writing mode that write request on described additional disk takes order to append, in the time that certain redirected write request is updated, new data order is appended in additional disk, and revising corresponding metadata is invalid by the old Data Identification in additional disk.

5. the method for reconstructing of a kind of solid-state disk array according to claim 1 and 2, it is characterized in that, described redirected metadata Hash table is made up of four fields, each field is followed successively by initial logical address S_LBA, the solid-state reel number DiskNo writing, band verification reel number ParityNo and the LBA (Logical Block Addressing) LBA on additional disk of band, and each list item in Hash table is used for recording the key message of a redirected write request.