JPH11143745A - Data management system and data management method - Google Patents

Abstract

(57) Abstract: Provided is a data management system for efficiently performing a process of repacking effective data blocks scattered in m stripes into n (m> n) stripes. A stripe management table for storing at least one of the number of valid blocks and the number of invalid blocks of each stripe on a magnetic disk device;
In accordance with storage of data in the stripe and deletion of data from the stripe, the stripe management table 11
Update the value of the number of valid blocks or the number of invalid blocks of the stripe stored in. Then, when rewriting effective data blocks scattered in m stripes into n (m> n) stripes, this stripe management table 11 can be used without performing actual disk access.
Selects the most suitable stripe to be subjected to the refilling process, based on at least one of the number of valid blocks and the number of invalid blocks held by.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data management system and a data management method which are preferably applied to a storage device in which writing and reading are performed in a stripe unit composed of a predetermined number of data blocks.
The effective data blocks scattered in the stripes are represented by n (m
The present invention relates to a data management system and a data management method for efficiently executing a process of refilling> n) stripes.

[0002]

2. Description of the Related Art In recent years, the spread of computers has been remarkable, and a large amount of data has been processed by computers. These data are stored in a storage device. In this storage device, writing and reading are performed in a stripe unit composed of a predetermined number of data blocks in order to store the data efficiently.

When data processing is executed by a computer, data is naturally updated.
With this data update, valid data blocks and invalid data blocks are mixed on the same stripe. Then, in the storage device, in order to efficiently store data, for example, refilling processing for refilling valid data blocks scattered in m stripes into n (m> n) stripes is required.

[0004]

As described above, in a storage device in which writing and reading are performed in a stripe unit composed of a predetermined number of data blocks, the effective data blocks scattered in m stripes are n (m> n). By performing the refilling process for refilling into ()) stripes, efficient data storage is ensured.

However, in this refilling process, when selecting a stripe to be refilled, it is necessary to determine whether there are valid blocks and invalid blocks in the stripe, and the number thereof. Conventionally, a stripe on a magnetic disk, for example, is actually accessed in order to obtain the presence / absence of valid and invalid blocks in the stripe. As a result, access to the magnetic disk may hinder input / output request processing of the host system.

The present invention has been made in view of such circumstances, and efficiently executes, for example, a process of rewriting effective data blocks scattered in m stripes into n (m> n) stripes. It is an object to provide a data management system and a data management method.

[0007]

In order to achieve the above-mentioned object, the present invention provides a stripe management table which holds at least one of the number of valid blocks and the number of invalid blocks of each stripe, and stores data in the stripe. And depending on the invalidation of the data on the stripe,
The value of the number of valid blocks or the number of invalid blocks of the stripe stored in the stripe management table is updated. For example, when refilling valid data blocks scattered in m stripes into n (m> n) stripes, the refilling is performed based on at least one of the number of valid blocks and the number of invalid blocks held in the stripe management table. A stripe to be processed is selected.

That is, according to the present invention, since there is no need to actually access a stripe on a magnetic disk in order to obtain the presence / absence of a valid block and an invalid block in a stripe, an input / output request processing of a host system is performed. Does not disturb.

[0009]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram showing a schematic configuration of a storage device according to an embodiment of the present invention. As shown in FIG. 1, a storage device 10 of this embodiment includes a stripe management table 11, a logical address conversion table 12, a control unit 13, a write buffer 14, a read buffer 15,
A refill buffer 16, a disk controller 17 and a magnetic disk device 18 are provided.

In the storage device 10, writing and reading are performed in a stripe unit composed of a predetermined number of data blocks. Writing and reading of the stripe to and from the magnetic disk device 18 are performed. The execution is controlled by a disk controller 17 to which each magnetic disk device 18 is connected.

FIG. 2 shows a logical address conversion table 12.
Is shown. The logical address conversion table 12 holds the number of a stripe in which a data block corresponding to a logical address is stored, and a block number indicating a position in the stripe. Further, the logical address conversion table 12 holds, as a time stamp, the time at which the block corresponding to the logical address was last updated. The time stamp may be stored in the stripe management table 11, which will be described later, as information in units of stripes, instead of in units of logical blocks.

FIG. 3 shows the format of the stripe management table 11. The stripe management table 11 stores the number of valid data blocks stored in a stripe and the magnetic disk device 1 where the stripe is located.
8 records the number of the disk controller 17 to which it is connected and a lock flag for exclusive control. When the number of valid blocks is 0, it means that the corresponding stripe is an empty stripe. If the number of valid blocks is a negative number, it means that the corresponding stripe does not exist on this storage device. The provision of the stripe management table 11 is a feature of the present invention.

Next, the stripe data structure of this embodiment will be described. FIG. 4 shows the data structure of the stripe of this embodiment. On the stripe,
A plurality of data blocks and tag information are stored. The tag information stores correspondence information between each data block on the stripe and the logical address and a time stamp, and occupies an area of one block size. The time stamp is information indicating the time at which this stripe was updated. By comparing this value with the time stamp of another stripe and the time stamp stored in the logical address conversion table, it is determined whether the data block contained in the stripe is valid or invalid.

Here, the outline of the write processing in the storage device 10 will be described together with the function of the write buffer 14. The write buffer 14 has the same capacity as one stripe. The control unit 13 divides the write data received from the host system 20 in units of a block size and temporarily stores the divided data in the write buffer 14. The control unit 13 records the logical address of the block and the position on the write buffer 14 (= the position on the stripe) in the tag information at the time of accumulation in the write buffer 14. Then, when the number of data blocks stored in the write buffer 14 becomes −1 (the number of blocks per stripe), the control unit 13 deletes the contents of the write buffer 14 with the tag information added to the free stripe on the magnetic disk 18. Write all at once.

The control unit 13 updates the number of valid blocks on the stripe management table 11 when writing to an empty stripe in the write buffer 14. This is as follows: (1) Updating the number of effective blocks of the stripe that has been written. (2) Updating the number of effective blocks of the stripe that contains the old data block. Consists of the following two processes.

The updating of the number of effective blocks of the stripe containing the old data block is performed in the following procedure. First, the logical address of each block stored in the write buffer 14 is used as a key, and the logical address conversion table 12 is referred to.
The stripe containing the old data of the logical address is specified. Next, the value of the number of effective blocks is decremented by one by referring to the stripe conversion table 11 using the stripe as a key. This process is performed for all data blocks on the write buffer 14. FIG. 5 shows a procedure for updating the number of valid blocks on the stripe management table 11.

Next, the outline of the read processing in the storage device 10 will be described together with the function of the read buffer 15. In this read processing, the logical address of the given host IO request is used as a key, and the logical address conversion table 12 is referenced to check the stripe in which the read target block is stored and the position in the stripe.
Then, after reading the stripe from the magnetic disk device 18 and storing it in the read buffer 15, the data of the corresponding block in the read buffer 15 is read and returned to the host system 20 as data.

The procedure of the refilling process will be described below. This refilling process includes three processes of “selection of a stripe to be refilled”, “relocation of valid blocks in a stripe”, and “writing to an empty stripe”. Then, all of these three processes require a process of checking the stripe for the presence or absence of an invalid block.

First, a conventional procedure for searching for an invalid block in a stripe will be described. Conventionally, first, a logical address and a time stamp value of each block in a stripe are obtained by referring to tag information in the stripe. Next, the logical address corresponding to each block obtained from the tag information is used as a key, and the logical address conversion table 12 is referenced to obtain a time stamp when the block was last updated. Then, the time stamp value is compared with the time stamp value of the tag information in the stripe, and if the values are the same, it is determined that the information of this block is valid. Conversely, if the stamp value of the tag information is an older value, it is determined that the block corresponding to this logical address is already invalid.

As described above, conventionally, in order to search for an invalid block in a stripe, it is necessary to access tag information in the stripe. That is, this means that it is necessary to perform a stripe read process. Therefore, in some cases, processing of the host IO request has been hindered.

Next, a procedure for searching for an invalid block and a valid block in this embodiment will be described. The storage device 10 of this embodiment includes a stripe management table 11 for holding valid blocks of a stripe. Therefore, to check for invalid blocks in a stripe,
The number of valid blocks may be compared with the number of blocks per stripe by referring to the stripe management table 11 using the corresponding stripe number. This is because if the number of valid blocks is smaller than the number of blocks per stripe, it means that there is an invalid block in the corresponding stripe.

As described above, since it is not necessary to perform a magnetic disk access to check whether or not there is invalid data in a block of a certain stripe, the processing of a host IO request is not hindered.

Next, an example of a method of selecting a stripe to be refilled will be described. The example shown here is an example of a simple method, and by using other selection methods, it is possible to obtain a more efficient combination of stripes to be refilled. In the following description, the number of blocks per stripe is represented by L. Therefore, in this embodiment, the number of data blocks that can be stored in the stripe is L-1. (1) Selection of stripes to be refilled (m TO
1) First, as one of the stripes to be refilled,
Select any one stripe. Then, this stripe number is registered in the refill target list. here,
The stripe management table 11 is searched, and a stripe other than an empty stripe and having the largest number of invalid blocks (the smallest number of valid blocks> 0) is selected.

Next, a stripe having an effective block whose sum with the total number of effective blocks of the stripe registered in the refill target list (hereinafter referred to as Σ) is closest to L−1 is searched. However, in this search, empty stripes and stripes in which all blocks are valid blocks are excluded. Then, the stripe obtained by the search is registered in the refill target list. Also, Σ <L−1
If it means that there is still an empty block in the stripe after refilling, the same procedure is repeated.

If the combination of Σ = L−1 is obtained, it means that the effective block in the stripe to be refilled can be relocated in one stripe by the refilling process. Then, a refilling process is performed on the stripes in the target list.

When a combination of Σ = L−1 cannot be obtained by such a procedure, one stripe cannot be created from a stripe to be refilled. In this case, the n stripes (m
>N> 1) is performed to create a stripe. (2) Selection of stripes to be refilled (m TO
n) In the procedure for creating n (m>n> 1) stripes from the m stripes, in the above-described procedure, Σ = (L
-1) The search is repeated until the number becomes a multiple of n times. (Relocation Processing Procedure) First, each stripe to be refilled selected in the above-described processing is read out to the refill buffer 16. Next, one of the read stripes is selected and used as a relocation stripe write buffer. When the number of stripes after the rearrangement becomes n, the refilling buffer 16 for the n stripes is used as a rearrangement stripe write buffer. Then, in the invalid block in the relocation stripe write buffer,
Valid blocks in other stripes to be refilled are sequentially overwritten and copied.

When the valid block is overwritten and copied to the invalid block, the logical address information in the tag information of the relocation stripe write buffer is updated. Further, for the moved block, the stripe number, block number and time stamp of the logical address conversion table 12 are updated. Then, after moving all the valid blocks, the time stamp of the tag information of the relocation stripe write buffer is updated with the newest value. (Writing from m stripes to n stripes) The stripe write buffer for rearranged stripes which has been rearranged in the above-described processing is written back to the magnetic disk device 18. The process at the time of writing back is performed by a normal host IO.
The request processing is the same as the processing when the write buffer 14 is filled. That is, an empty stripe is selected, and the contents of the write buffer for relocation stripe are written to this empty stripe. At the time of write-back, the number of effective blocks of each of the empty stripes and the old data blocks that have been written in the stripe management table 11 is updated.

In selecting an empty stripe, the stripe management table 11 is searched, and a stripe to be written back is selected from the empty stripes. When rearrangement is performed on n stripes, a plurality of empty stripes to be rewritten are selected so as to belong to different disk controllers 17 from each other. As a result, it is possible to perform write processing on the stripe in parallel.

Here, writing back to the empty stripe is performed as follows.
Even if something goes wrong during the write-back operation,
This is to enable restoration from the original stripe information. Therefore, if there is a means for guaranteeing the write-back data and the completion of the write-back when a failure occurs, a stripe to be refilled may be used instead of an empty stripe.

By the way, the refilling process described so far is to actually perform the reading of the stripe existing on the magnetic disk device 18 to perform the refilling process. Therefore, this process involves at least two stripe reading processes on the magnetic disk device 18 and one or more writing processes to the magnetic disk device 18. Such processing involving magnetic disk access may cause a state in which a host IO request is made to wait.

Hereinafter, as a method for solving this problem,
Various buffers of the storage device 10 (the write buffer 14,
A method for making valid data blocks existing on the read buffer 15) a target of the refilling process will be described. Here, a case will be described in which one of the target stripes for the refilling process is the stripe data stored in the read buffer 15.

In this case, the “arbitrary stripe” in the above-described procedure for selecting a combination of stripes to be refilled is registered as a “read buffer 15” in the refill target list. Thereafter, similarly, a stripe having an effective block in which the number (有効) of effective blocks in the stripe registered in the refill target list is equal to the number L-1 of data blocks of the stripe is selected. Then, the selected stripe (excluding the read buffer) is read into the refill buffer 16.

Next, a valid block is searched for from the stripe data in the refill buffer 16, and the valid block is relocated to an invalid block on the read buffer 15.
The procedure and processing at the time of rearrangement are as described above. Then, after all the relocation processes are completed, the time stamp value in the read buffer 15 is updated with the latest value, the data in the read buffer 14 is written to an empty stripe, and the number of valid blocks in the stripe management table 11 is updated. I do.

However, the contents of the write buffer 14 must not be written to the magnetic disk device 18 before the stripe on the magnetic disk device 18 is updated by the refilling process. Because write buffer 1
If the value of the time stamp of the tag information on the read buffer 4 is older than the time stamp value of the tag information on the read buffer 15 updated by the refilling process, the write data of the host IO request is This is because it is regarded as old information and will be lost. A method of preventing this is to suspend the writing process of the write buffer 14 to the magnetic disk device 18 until the refilling process is completed, or advance the time stamp of the write buffer 14 in advance.

Note that one of the stripes to be refilled is
In the case where one is the stripe data stored in the write buffer 14, similarly to the case of the read buffer 15, the information updated by the host IO request is not lost with the movement of the effective data of the refilling process. You need to do that.

By the way, if such refilling processing is performed, each buffer becomes busy, and there is a possibility that host IO requests may be obstructed. As measures against this, (1) the write buffer 14 or the read buffer 1
In the case where No. 5 is to be subjected to the refilling process, it is assumed that the host IO request has not been generated for a certain time. (2) When a plurality of buffers are provided and refilling processing is performed on the buffers, another buffer is used. And so on.

In the case where a plurality of stripes are read out from a plurality of stripes and the refilling process is performed, including the case where various buffers are used, the stripes to be refilled exist under different disk controllers 17 from each other. It is preferable to select as follows. As a result, the stripes to be refilled can be read out in parallel, and the time required for stripe refilling processing can be significantly reduced. In such a stripe selection, in the above-described stripe selection procedure, a condition that “the disk belongs to a disk controller 17 different from the disk controller 17 to which the stripe registered in the refill target list belongs” is added as a stripe selection condition. This can be achieved by: The disk controller 17 to which each stripe belongs can be determined by referring to the controller number in the stripe management table 11.

As described above, the storage device 10 of this embodiment
According to this, it is possible to greatly reduce the number of times of access to the magnetic disk caused by the refilling process, and to greatly reduce the time required for accessing the magnetic disk. Note that this data management method is based on the host system 2
It may be controlled on the 0 side.

[0039]

As described above in detail, according to the present invention, the presence / absence of valid blocks and invalid blocks in a stripe on a magnetic disk device, and the numbers thereof are examined without access to the magnetic disk. Therefore, the stripe refilling process does not hinder access to the magnetic disk by input / output request processing from the host system.

Further, by making the read buffer and the write buffer one of the stripe data to be refilled, it is possible to suppress the occurrence of a magnetic disk access performed only for the refill processing.

Further, even when performing magnetic disk access accompanying the refilling process, it is possible to access in parallel by allocating the disk to stripes belonging to different disk controllers.

[Brief description of the drawings]

FIG. 1 is a diagram showing a schematic configuration of a storage device according to an embodiment of the present invention.

FIG. 2 is an exemplary view showing the format of a logical address conversion table according to the embodiment;

FIG. 3 is an exemplary view showing the format of a stripe management table according to the embodiment;

FIG. 4 is an exemplary view showing a data structure of a stripe according to the embodiment;

FIG. 5 is an exemplary flowchart showing the procedure of processing for updating the number of valid blocks on the stripe management table according to the embodiment.

[Explanation of symbols]

10: storage device, 11: stripe management table, 12
... Logical address conversion table, 13 control unit, 14 write buffer, 15 read buffer, 16 refill buffer, 17 disk controller, 18 magnetic disk drive, 20 host system.

Claims (13)

[Claims] 1. A data management system applied to a storage device in which writing and reading are performed in a stripe unit composed of a predetermined number of data blocks, wherein the stripe holding at least one of the number of valid blocks and the number of invalid blocks of each of the stripes A management table, and table updating means for updating a value of the number of valid blocks or the number of invalid blocks of the stripe stored in the stripe management table in accordance with storage of data in the stripe and invalidation of data on the stripe. A data management system comprising: 2. The data management system according to claim 1, further comprising a search unit that searches for an empty block in which the number of valid blocks held in the stripe management table is 0. 3. A plurality of controllers for controlling writing and reading of a predetermined number of stripes are provided, and the stripe management table further holds identification data for identifying a controller to which each of the stripes belongs. Data management system. 4. The apparatus according to claim 1, further comprising: refilling means for refilling the effective data blocks scattered in the m stripes into n (m> n) stripes, wherein the refilling means comprises the number of effective blocks held in the stripe management table. 4. The data management system according to claim 1, wherein a stripe to be refilled is selected based on at least one of the number of invalid blocks and the number of invalid blocks. 5. The data management system according to claim 4, wherein said refilling unit regards a write buffer for writing data in said storage device as one of stripes and selects it as a stripe. 6. The refilling means considers that a write buffer for writing data to the storage device is regarded as one of stripes and is to be selected when a write request does not occur for a predetermined time. 5. The data management system according to claim 4, wherein: 7. A plurality of said write buffers are provided,
7. The data according to claim 5, further comprising buffer control means for accumulating write data in another write buffer while one of the write buffers is subjected to refill processing by said refill means. Management system. 8. The data management system according to claim 4, wherein said refilling unit regards a read buffer for reading data from said storage device as one of stripes and selects it as a stripe. 9. The refilling unit according to claim 1, wherein when a read request does not occur for a predetermined time, a read buffer for reading data from the storage device is regarded as one of the stripes and is selected. 5. The data management system according to claim 4, wherein: 10. A buffer control means, wherein a plurality of said read buffers are provided, and a buffer control means for storing a stripe including read data in another read buffer while any one of the read buffers is subjected to refill processing by said refill means. The data management system according to claim 8, further comprising: 11. The data management system according to claim 4, wherein said refilling means selects stripes to be subjected to said refilling processing such that stripes belonging to different controllers are paired. 12. The data management system according to claim 4, further comprising: write control means for performing write control so that a plurality of stripes refilled by said refill means belong to different controllers. 13. A data management method applied to a storage device in which writing and reading are performed in a stripe unit composed of a predetermined number of data blocks, wherein at least one of the number of valid blocks and the number of invalid blocks of each of the stripes is held. And updating the held value of the number of valid blocks or the number of invalid blocks of the stripe in accordance with storage of data in the stripe and invalidation of data on the stripe. Data management method.