JP3431581B2 - Disk control system and data relocation method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a disk control system and a data relocation method, and more particularly to a disk control system using a log structured writing method and a data relocation method applied to the system.

[0002]

2. Description of the Related Art In recent years, RAID (Redundant Array) has been used from the viewpoint of speeding up a disk and improving reliability.
Computer systems equipped with of Independent Disks) technology have become mainstream. RAID
There are 0, 1, 10, and 5 types of technology.

Recently, as a technique for greatly improving the write performance of such a RAID, a log structured file system (LSFS: Log-Structure) is used.
edFile System) technology has been considered.
The log-structured file system is a method of performance improvement that utilizes the characteristic of the disk device that "sequential access is very fast compared to random access". The performance degradation during random access is due to disk seek and rotation waiting, and is the part that we want to eliminate if possible. Therefore, in the log structured file system technology, there is a method of using a buffer for accumulating write data, converting write data of a plurality of small blocks into one large chunk of data, and performing “lump writing”. Used. As a result, random writing can be changed to sequential writing, and the writing performance can be improved.

[0004]

However, when the log structured file system technology is used, the data is arranged and written in the order of write request. Therefore, when random write occurs, data whose logical addresses are apart from each other are continuously written in one physical stripe area. As a result, on the file system, data blocks having consecutive logical addresses are written to different physically striped areas that are physically separated. In this way, sequential reading of data physically distributed and stored is physically random and inefficient.

In order to correct such a state, it is necessary to regularly rearrange the data so that the data are physically continuous. However, the process of rearranging the data itself involves a lot of I / O processing, which imposes a heavy load on the system.

The present invention has been made in view of the above circumstances, and enables efficient data relocation processing,
An object of the present invention is to provide a disk control system and a data rearrangement method capable of significantly improving read performance with a small load.

[0007]

In order to solve the above-mentioned problems, the present invention is a disk control system of a log structured write system, wherein a disk array composed of a plurality of disk devices and a write request are made. A data buffer for accumulating a plurality of data blocks, and write data for one physical stripe of the disk array is generated from the plurality of data blocks accumulated in the data buffer, and collectively written in a predetermined physical stripe of the disk array. Data writing means for writing the data, a plurality of logical block numbers forming a logical stripe of the disk array, and a physical block number indicating a physical position on the disk array where a data block designated by the logical block number exists. Address conversion to store correspondence for each logical stripe Table and the address conversion table, the ratio of the number of valid logical blocks in the logical stripe and / or the ratio of consecutive adjacent physical block numbers is calculated, and the result satisfies a predetermined condition. In this case, the relocation target determining unit that determines the logical stripe to be relocated, and the data block of the logical stripe whose relocation is determined by the relocation target determining unit are read from the disk array to the data buffer, and The data writing means comprises data rearrangement means for writing the data in the empty physical stripe area of the disk array, and address conversion table rewriting means for rewriting the physical block number of the address conversion table to the physical block number of the empty physical stripe. It is characterized by

In this disk control system, the degree of physical distribution of logical blocks is checked for each logical stripe by referring to the address conversion table. That is, for each logical stripe, the ratio of the number of valid logical blocks in a plurality of logical blocks constituting the logical stripe and / or the ratio of consecutive physical block numbers of adjacent logical blocks is calculated, and the rearrangement is performed based on the calculated ratio. The target logical stripe is determined. A logical stripe having a small ratio of the number of effective logical blocks cannot be expected to have a sufficient effect even if it is rearranged, and therefore such a logical stripe is preferably excluded from the rearrangement target. It also contains some valid logic blocks, and
Logical stripes with low physical block number continuity are inefficient when performing sequential reads, but the efficiency of sequential reads can be improved by making the data blocks in such logical stripes reallocation targets. .

The present invention is also a log-structured write type disk control system, comprising a disk array comprising a plurality of disk devices, a data buffer for accumulating a plurality of write-requested data blocks, and a plurality of data blocks stored in the data buffer to generate a physical stripe write data of the disk array, and a data writing means for writing collectively a predetermined physical stripe of the disk array, the
Multiple Theories Constituting Logical Address Space of Disk Array
Logical block number and the logical block number
Physical location on the disk array where the data block resides
Address that stores the correspondence with the physical block number
Address conversion table and the logical address of the disk array.
Space for one physical stripe worth of data blocks
For each logical stripe separated by
For multiple logical blocks contained within
Counter means for counting the number of data reads performed, relocation target determining means for determining a logical stripe having the number of data reads of a predetermined value or more as a logical stripe to be relocated, by referring to the counter means, A data relocation unit that reads the data block of the logical stripe whose relocation is determined by the relocation target determination unit from the disk array to the data buffer, and writes the data block in an empty physical stripe area of the disk array by the data writing unit; An address conversion table rewriting unit for rewriting the physical block number of the address conversion table to the physical block number of the empty physical stripe.

As described above, by performing data rearrangement on a logical stripe having a large number of data read times, it becomes possible to make the data blocks in the logical stripe that are frequently read physically continuous. It is possible to efficiently improve the lead performance.

The present invention is also a log structured write type disk control system, comprising a disk array composed of a plurality of disk devices, a data buffer for accumulating a plurality of write-requested data blocks, and a plurality of data blocks stored in the data buffer to generate a physical stripe write data of the disk array, and a data writing means for writing collectively a predetermined physical stripe of the disk array, the
Multiple Theories Constituting Logical Address Space of Disk Array
Logical block number and the logical block number
Physical location on the disk array where the data block resides
Address that stores the correspondence with the physical block number
Address conversion table and the logical address of the disk array.
Space for one physical stripe worth of data blocks
For each logical stripe separated by
For multiple logical blocks contained within
Counter means for counting the number of times of data writing, relocation target determining means for determining a logical stripe having the data writing frequency of a predetermined value or more as a logical stripe to be relocated by referring to the counter means, and the relocation Data relocation means for reading the data block of the logical stripe whose relocation is decided by the object deciding means from the disk array to the data buffer, and writing it to an empty physical stripe area of the disk array by the data writing means, and the address. And an address conversion table rewriting unit for rewriting the physical block number of the conversion table to the physical block number of the empty physical stripe.

A logical stripe may become physically discontinuous when a logical block forming the logical stripe is written. Therefore, by performing the process for determining the logical stripe to be relocated to the logical stripes for which the number of data writes is equal to or greater than the predetermined value,
It is possible to efficiently improve the performance of sequential access.

[0013]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows the configuration of a computer system to which a disk control system according to an embodiment of the present invention is applied. This computer system is a server (PC
It is used as a server) and has a plurality of CPUs 1.
1 can be mounted. Each of these CPUs 11 is connected to the bridge 12 via the processor bus 1 as shown. Bridge 12 is processor bus 1 and PCI
It is a bridge LSI for connecting the bus 2 bidirectionally, and a memory controller for controlling the main memory 13 is also incorporated therein. An operating system (OS), application programs to be executed, drivers, and the like are loaded in the main memory 13.

The PCI bus 2 is connected to the RAI as shown in the figure.
The D controller 16 and the non-volatile memory 17 are connected. The disk array 18 controlled by the RAID controller 16 is used for recording various user data and the like.

Under the control of the RAID controller 16, the disk array 18 functions as a disk array having a RAID5 configuration, for example. In this case, the disk array 18
Is N + 1 units (here, one disk unit for parity storage is added to N disk units for data storage).
It is composed of five disk devices (DISK0 to DISK4). These N + 1 disk devices are grouped and used as a single logical disk drive.

As shown in the figure, the grouped disk devices have data (DATA) and its parity (PAR).
PHY) and a physical stripe (parity group) each composed of TY), and the parity position is sequentially moved between N + 1 disk devices for each physical stripe. For example, in the physical stripe S0, DISK0-D
The parity (PA) of the data (DATA) group on the stripe unit allocated to the same position of each ISK3
RITY) is recorded on the corresponding stripe unit of DISK4. In the next physical stripe S1, the parity (PARTY) corresponding to the data of the stripe S1 is recorded on the corresponding stripe unit of DISK3. By thus distributing the parity in N + 1 disk devices in units of physical stripes, it is possible to prevent the concentration of access to the parity disk.

The non-volatile memory 17 is a working storage device used for implementing a log structured file system (LSFS), and is a battery backed up DRA.
It is composed of M or flash EEPROM. Log Structured File System (LSFS) is R
It is used to improve the writing performance for the disk array 18 having the AID5 configuration. In the present embodiment, the file system of the OS is not improved, and the log structured file system (LS) is configured by the driver program and the non-volatile memory 17 incorporated in the OS.
FS) has been realized. Here, the principle of write control of the disk array 18 by the log structured file system (LSFS) will be described with reference to FIG.

Log Structured File System (LSFS)
In the write method using the technology, the write data position is not sequentially determined according to the logical address write requested by the host (OS file system in this embodiment), but the write data is sequentially accumulated in the order in which the write request is made by the host. By doing so, a large data block composed of a plurality of blocks of write data is configured, and the large data block is collectively "written" in order from the top in the empty area of the disk array 18. The unit of "bulk writing" is a physical stripe. That is, one empty physical stripe is generated every "bulk writing", and data blocks for one physical stripe are sequentially written therein. As a result, random access can be converted into sequential access, and the write performance can be greatly improved.

In FIG. 2, the block data size of the write data sent via the OS file system is 2 KB, and the data size of one stripe unit is 64 K.
In this example, the data size of one physical stripe (parity group) is 256 KB (= 64 KB × 4). The 2 KB write data block is acquired by the RAID acceleration driver 100 used by being incorporated in the OS and is stored in the non-volatile memory 17. RAID
The speed-up driver 100 is a driver program for realizing a log structured file system (LSFS).

Basically, at the time when a data block of 256 KB (2 KB × 128 data blocks) is accumulated in the non-volatile memory 17, one physical unit of the disk array 18 is controlled under the control of the RAID acceleration driver 100. Writing to stripes is performed at once. In this case, since the RAID controller 16 can generate the parity from only the write data block of 256 KB, it is not necessary to perform processing such as reading old data for the parity calculation, and the well-known write penalty of RAID 5 can be reduced. it can.

The non-volatile memory 17 includes a log memory (collective writing buffer) 17 as shown in FIG.
1, an address conversion table 172, a logical stripe management table 173, etc. are mounted, and access control is performed by a memory control unit (not shown).

The log memory 171 is a data buffer for accumulating write data blocks. When the write data blocks for one physical stripe are accumulated in the log memory 171, batch writing to the disk array 18 is started. It That is, the physical stripe is a unit of “collective writing”, and is composed of a continuous continuous area on the partition created for LSFS in the entire storage area of the disk array 18. Each physical stripe has a size that is an integral multiple of the stripe unit managed by the RAID controller 16.

The address conversion table 172 is the LSFS.
Correspondence between each of a plurality of logical block numbers forming a logical address space of a partition used as a partition and a physical block number indicating a physical position on the disk array 18 in which a data block designated by these logical block numbers exists. Is stored.
Here, the logical block number is the number of the logical block on the disk partition as viewed from the OS file system. The block number when the OS file system requests access is a logical block number, which is a virtual logical address. This logical block number is associated with the physical block number (physical block number on the disk partition) by the address conversion table 172. The byte offset value from the head position of the disk partition is obtained by the physical block number × block size (bytes).

The address conversion table 172 has a plurality of entries corresponding to respective logical block numbers. When writing a new data block, the physical block number (physical address) in which the data block is actually written is registered in the entry corresponding to the write-requested logical block number. When reading a data block, the physical block number is checked from the entry corresponding to the logical block number requested to be read,
The data block is read from the physical position on the disk partition designated by the physical block number (physical block number x block size).

The logical stripe management table 173 is for managing information about each logical stripe, and the information of each logical stripe is used for processing for data relocation. Here, the logical stripe is a set of logical blocks in which the logical address space of the disk partition is divided into data blocks of one physical stripe from the beginning. For example, when one physical stripe includes 10 physical blocks,
One logical stripe will also include 10 logical blocks.

As described above, in the writing method of this embodiment, random writes are collectively written in the physical stripe as one continuous area. As a result, even if the randomly written area is continuous data in the logical address space on the file system of the OS, it is included in the physically separated physical stripes. In this way, the sequential read for the data written in the different physical stripes is physically a random read and is inefficient. In order to correct such a state, it is necessary to rearrange the data blocks periodically so that the data are physically arranged continuously. This data rearrangement process, that is, "data rearrangement for the purpose of efficiently performing sequential read" will be referred to as repacking process here.

The control method of the repacking process according to the first embodiment will be specifically described below with reference to FIGS. 3 and 4. FIG. 3 shows an example of the address conversion table 172, and FIG. 4 is a flowchart of the repack control process.

The address conversion table 172 of FIG. 3 assumes that the physical stripe (also the logical stripe) consists of 10 data blocks. The address conversion table 172 includes a large number of entries with the logical block number as an index, and the physical block number is registered in each entry. "-" Indicates an invalid block (a block in which data has not been written yet).

In the example of the address conversion table 172, one logical stripe is formed for every 10 logical blocks from the top. For example, logical stripe 0 consists of 10 logical blocks with logical block numbers 0 to 9,
Logical stripe 1 is 1 of logical block numbers 10 to 19
It consists of 0 logical blocks. That is, when one physical stripe and one logical stripe each consist of 10 data blocks, the following relationship holds: logical stripe number = logical block number / 10 physical stripe number = physical block number / 10. However, in the above, “/” indicates division of an integer (rounding down after the decimal point).

Focusing on the logical stripe 0, the data block having the logical block number 0 is the physical block number 30.
Similarly, the data block with the logical block number 1 is stored in (the first data block of the physical stripe 3), and the data block with the logical block number 2 is the physical block with the physical block number 31 (the second data block of the physical stripe 3). It can be seen that the data block with the number 32 (the third data block of the physical stripe 3) and the data block with the logical block number 3 are stored with the physical block number 33 (the fourth data block of the physical stripe 3). Further, the data blocks corresponding to the logical block numbers 6, 7, and 8 are physical block numbers 133, 134, and 2, respectively.
37, respectively. Logical block number 4,
Reference numerals 5 and 9 are invalid blocks.

In this way, the address conversion table 172
By referring to, it is possible to detect the degree of physical distribution of the logical blocks constituting each logical stripe. Therefore, in the present embodiment, by referring to the address conversion table 172 and checking the physical block numbers of the logical blocks forming each logical stripe, the ratio of the number of effective blocks (=
α), the ratio (= β) in which the physical block numbers of adjacent logical blocks are continuous is calculated for each logical stripe. Whether or not the logical stripe is to be repacked is determined by the values of α and β.

(Calculation Example of α and β) Here, a calculation example of α and β will be described. In FIG. 3, logical stripe 0
With respect to, since the number of effective blocks is 7, α of the logical stripe 0 is calculated by α = 7/10 × 100 = 70%. The value of α indicates the capacity usage rate of the target logical stripe. It can be judged that the smaller the value is, the more blocks are not written, and that the blocks should be excluded from the repacking target (there are few blocks to be rearranged even if repacking and the effect cannot be expected).

In the logical stripe 0, the physical block numbers of the logical blocks 0, 1, 2, 3 are consecutive,
The physical block numbers of the logical blocks 6 and 7 are consecutive. Therefore, among the nine block boundaries, four boundaries are physically continuous. Therefore, β of the logical stripe 0 is calculated by β = 4/9 × 100≈44%.

Similarly, α = 80% and β = 22% for the logical stripe 1, α = 20% and β = 0% for the logical stripe 2, and α for the logical stripe 3.
= 90% and β = 44%.

(Judgment of repacking) Whether or not repacking should be performed for each logical stripe by using "repack target when α is larger than a fixed value A and β is a fixed value B or less" (condition 1) To determine. Where A = 50%,
B = 50%.

In this case, since the logical stripe 2 has a small value of α, it does not meet the above condition and it is judged that repacking is unnecessary. This is because "the number of effective blocks is small and the effect of sorting cannot be expected". The logical stripes 0, 1, 3 meet the above conditions, and are determined to be repacked. It is targeted for repacking because "the number of effective blocks is sufficient and continuity is low".

Actually, α and β of each logical stripe change as the repacking process is performed. Therefore, in this embodiment, the repacking process is sequentially performed for each logical stripe determined to be the repacking target, and α and β for the next logical stripe are calculated after the repacking process is completed.

(Procedure of Repack Control Processing) The procedure of the repack control processing will be described below with reference to the flowchart of FIG. This repack control process is performed by the RAID acceleration driver 100.

First, the address conversion table 172 is referred to in the order of indexes to check the value of the corresponding physical block number (step S11). Then, the logical stripes with smaller values are selected in order (step S12), and the following "repack target stripe selection processing & repack processing (* 1)" is executed for the selected logical stripes.

In the repacking target stripe selection processing and repacking processing, first, α and β relating to the first selected logical stripe are calculated (step S13). The value of the number of valid blocks in each logical stripe is the number of read references,
It is managed for each logical stripe in the logical stripe management table 173 together with information such as the number of writes. The logical stripe management table 173, together with the address conversion table 172, is saved in an area different from the data area in the disk partition for LSFS when the computer system is shut down, and is nonvolatile from the disk partition when the computer system is booted. It is loaded into the memory 17 and used. The value of α is
It can be calculated based on the number of valid blocks in the logical stripe management table 173. Of course, as described above, by registering the information "-" indicating the invalid block in the address conversion table 172, α can be easily obtained only by searching the entry of the address conversion table 172.

When "α is equal to or smaller than the constant value A" (YES in step S14), the current logical stripe is excluded from the repacking target. On the other hand, when “α is larger than the constant value A” (NO in step S14), β is the constant value B.
It is determined whether or not the following (step S15).
When “α is larger than the constant value A and β is equal to or smaller than the constant value B” (YES in step S15), the current logical stripe is determined as a repack target, and the repack process is executed on the logical stripe. (Step S
16). The repacking process is a process of rearranging data blocks having consecutive logical block numbers so as to be physically consecutive. In this repacking process, while the data block of the logical stripe to be repacked is read from the disk array 18 to the log memory 171, an empty physical stripe is prepared on the partition of the disk array 18 and read to the log memory 171 there. Data blocks that have been written are collectively written in the order of logical block numbers.

Next, it is judged whether the processing has been completed for all the logical stripes (step S1).
7) The repacking target stripe selection process and the repacking process are sequentially performed in the index order until there are no unprocessed logical stripes.

The repack control process described above may be automatically executed, for example, at regular intervals, or may be executed in response to an explicit operation instruction from the system administrator.

6 and 7 show an example of the result of repacking the logical stripe 3 shown in FIG. FIG. 6 shows the contents of the address conversion table 172 before and after repacking regarding the logical stripe 3.
Further, FIG. 7 shows how data blocks are arranged on the physical stripe before and after repacking regarding the logical stripe 3.

Before repacking, logical stripe 3
9 valid blocks (logical block numbers 30, 3
, ... 38) are distributed and stored in three physical stripes (physical stripes 6, 11, 15). Note that reference numerals 30 to 38 in FIG. 7 mean data of the logical block number. If the logical blocks 30 to 38 are continuously read in this state, the three physical stripes 6, 11, 15 which are distant from each other are actually accessed, which is inefficient. Repacking is performed on this logical stripe 3. Now, an empty physical stripe 1 on the disk array 18
Assuming that 0 is selected, the data of all valid blocks of the logical stripe 3 will have three physical stripes 6,1.
1, 15 are read to the log memory 171 and are collectively written in the physical stripe 10. in this case,
In the physical stripe 10, all nine effective blocks are arranged in order of logical block number (30, 31, ... 38). In accordance with this, in the address conversion table 172, the physical block numbers of the nine effective blocks (logical block numbers 30, 31, ... 38) in the logical stripe 3 are updated (100, 10).
1, ... 108). By this repacking process, when the logical blocks 30 to 38 are continuously read, only the access within the physical stripe 10 is made and efficient access is realized. As a result, the sequential read performance can be improved.

In this way, a new physical stripe is generated for each logical stripe to be repacked, and the data blocks included in the logical stripe to be repacked are arranged in the logical block number order to repeat the repacking process. As a result, the physical data arrangement comes closer to the logical data arrangement. Further, if the logical stripe units are arranged so as to be physically continuous, the continuity of the data blocks can be realized even at the physical stripe boundary. In the repacking process, it is most preferable to arrange the blocks in the order of the logical block numbers, but it goes without saying that the read performance can be improved by accommodating all the data blocks belonging to the logical stripe to be repacked in the same physical stripe.

It is also possible to add a condition that "if β is equal to or less than a predetermined value, subject to repacking", and select a logical stripe subject to repacking only from β.

(Repack Selection Processing Based on Read Reference Count) Next, with reference to FIGS. 8 and 9, a second embodiment of the repack processing will be described. Here, the read reference count is managed for each logical stripe, and the "repack target stripe selection process" is preferentially executed for the logical stripes for which the read reference count has exceeded a certain value. The value of the read reference count of each logical stripe means the read reference count after the last repacking target stripe selection process for that logical stripe is performed. Therefore, the value of the read reference count of each logical stripe is cleared to 0 every time the repacking target stripe selection process of the logical stripe is performed. In the case of a logical stripe that has never been subjected to repack selection processing, the read reference count has been the number of times since the partition was created.

FIG. 8 shows a RAID when reading a data block.
7 is a flowchart showing a processing procedure executed by the speed-up driver 100. Here, it is assumed that a read reference counter is prepared for each logical stripe, and that the read reference counter is used to manage the read reference count of each logical stripe. Referenced_Strip
e_Table [] is a table of "read reference counter", which has a logical stripe number as an index and a "read reference counter value" as a value. Initialized with 0 when a partition is created.

As shown in FIG. 8, at the time of reading a data block, first, the logical stripe number L to which it belongs is obtained from the logical block number requested to be read by the file system of the OS (step S21). If the logical block number of the read target is Vn and the total number of blocks in each logical stripe is Nb, L is given by L = Vn / Nb.

Next, Referenced_Str
“Read reference counter” of the logical stripe L to which the target logical block belongs, as pe_Table [L] ++
Is incremented (step S22). After that,
Normal read processing is performed (steps S23 and S2).
4). That is, the physical block number Pn corresponding to the logical block number Vn is obtained by referring to the address conversion table 172 (step S23), and the physical position on the disk partition designated by the disk address given by Pn × block size. Then, the corresponding data block is read out through the RAID controller 16 (step S24).

FIG. 9 is a flowchart of the processing executed by the RAID high speed driver 100 during the repack processing. In FIG. 4, the contents of the address conversion table 172 are referred to in the order of the index.
“Repacking target stripe selection process” using α and β described above for a logical stripe having a large number of read references, using erased_Stripe_Table [].
Is done.

That is, first, the logical stripe numbers are sorted in descending order by the value of the "read reference counter", and the process of selecting the stripes to be repacked from the logical stripe with the largest read reference counter value is performed (step S31). . In step S31, the new array A
[] Is created. Each entry in the array A [] has two members, Logical_Stripe_ID, and R.
It consists of the Efence_Count and is written in C language as A []. Logical_Stripe_ID A []. It is represented as Reference_Count. A []. Logical_Str
ipe_ID is a logical stripe number (Reference)
ed_Stripe_Table [] index), and A []. Reference
_Count indicates the number of read references. Referen
By sorting the values registered in ced_Stripe_Table [], "read reference counter"
The value and the number of the logical stripe are set in the two members of the array A in descending order of the value of.

Thereafter, the entries are referred to in the order of registration in the array A [], and the read reference counter "A [] .R.
As long as there is a logical stripe in which the "reference_Count" is equal to or more than the constant value C, the "repack target stripe selection process & repack process (process of * 1 in FIG. 4)" is performed. That is, first, with n = 0, the logical stripe (L0 = A [0].
Pay attention to _Stripe_ID (steps S32, S
33), the read reference count (A [0] .Refe
It is determined whether or not (rence_Count) is less than or equal to a constant value C (step S34). If the read reference count is larger than the fixed value C, the above-mentioned repack target stripe selection processing and repack processing based on α and β is executed (step S35). The read reference counter of the relevant logical stripe is cleared to 0 during the repacking target stripe selection processing regardless of whether it is selected as the repacking processing target (step S36).

After that, as n + 1, the logical stripe (L1 = A [1] .Logi having the second largest number of read references is set.
Pay attention to cal_Stripe_ID) (step S3
7, S38, S33), the read reference counter (A
[1]. Reference_Count) is a constant value C
It is determined whether or not the following (step S34).

Read reference counter "A [n] .Refe"
When a logical stripe whose “rance_Count” is equal to or smaller than the constant value C appears (YES in step S34), the process ends. Therefore, it is possible to give priority to the repacking target by giving priority to a logical stripe having a large number of read references. In addition, the read reference counter "A [] .Reference.
Since it is sufficient to perform the repacking target stripe selection process only for the logical stripes whose _Count is equal to or greater than the constant value C, it is possible to reduce the load required for the process.

(Repack Selection Process Based on Number of Writes) Next, with reference to FIGS. 10 and 11, a third embodiment of the repack process will be described. Here, the number of writes of the logical blocks belonging to each logical stripe is managed, and the “repack target stripe selection process” is preferentially executed for the logical stripes for which the number of writes exceeds a certain value. The value of the write count of each logical stripe means the write count after the last repacking target stripe selection process for that logical stripe is performed. Therefore, the value of the number of times of writing of each logical stripe is cleared to 0 every time the repacking target stripe selection process of the logical stripe is performed. It should be noted that in the case of a logical stripe that has never been the target of repack selection processing, the number of times of writing is "the number of times of writing after the partition is generated".

A logical stripe may become physically discontinuous when a logical block constituting it is written. This is because in LSFS, additional write processing is performed in physical stripe units, and the old data block on another physical stripe corresponding to the data block included in the newly written physical stripe is invalidated. Therefore, it is considered that the more the logical stripes are written, the higher the possibility that the physical positions where the logical blocks exist are dispersed. Therefore,
Efficient repacking processing can be realized by preferentially making a logical stripe with a large number of writing times a candidate for repacking processing.

FIG. 10 is a flowchart showing the processing procedure executed by the RAID acceleration driver 100 when writing data. Here, it is assumed that a write count counter is prepared for each logical stripe and the write count of each logical stripe is managed using the write count counter. Modified_Stripe_T
The table [table] is a write counter, and the logical stripe number is used as an index.
It has a "write count counter value" as a value. Initialized with 0 when a partition is created.

As shown in FIG. 10, when one physical stripe worth of data blocks are accumulated in the log memory 171,
The data blocks for one physical stripe are collectively written to the physical stripe to be written (step S41). Next, for each written data block, the logical stripe number L to which it belongs is obtained (step S42). The logical block number of each data block is V
n, and the total number of blocks in the logical stripe is Nb,
The logical stripe number L is given by L = Vn / Nb.

After this, Modified_Stripe
As _Table [L] ++, the “write number counter” of the logical stripe to which the target logical block belongs is incremented (step S43). After that, as a part of the normal write processing, the physical block number Pn actually written is set to the entry of the address conversion table 172 corresponding to the logical block number Vn for each written data block ( Step S44). Note that steps S41 to S44 described above are executed for all blocks in the physical stripe.

FIG. 11 is a flowchart of the processing executed by the RAID high speed driver 100 during the repack processing. In FIG. 4, the contents of the address conversion table 172 are referred to in the order of the index.
The "repack target stripe selection process" using α and β described above is performed for a logical stripe that has been written many times using defined_Stripe_Table [].

That is, first, the logical stripe numbers are sorted in descending order by the value of the "write count counter", and the repacking target stripe selection process is performed from the logical stripe with the largest write count counter value (step S51). In step S51, the new array A
[] Is created. Each entry in array A [] has two members, Logical_Stripe_ID, and M.
composed of modified_Count, and written in C language, A []. Logical_Stripe_ID A []. It is represented as Modified_Count. A []. Logical_Str
ipe_ID is a logical stripe number (Modified
_Stripe_Table [] index), and A []. Modified_Co
unt indicates the number of times of writing. Modified_St
By sorting the values registered in the ripe_Table [], the value and the number of the logical stripe are set in the two members of the array A in descending order of the value of the "write number counter".

Thereafter, the entries are referred to in the order of registration in the array A [], and the write count counter "A [].
As long as there is a logical stripe whose Modified_Count is equal to or larger than the constant value D, the “repack target stripe selection process & repack process (process * 1 in FIG. 4)” is performed. That is, first, with n = 0, the logical stripe (L0 = A [0] .Logical_
Pay attention to the Stripe_ID (steps S52 and S5)
3), the write counter (A [0] .Modi
It is determined whether (fied_Count) is equal to or less than the constant value D (step S54). When the number of times of writing is larger than the constant value D, the above-mentioned repacking target stripe selection processing & repacking processing based on α and β is executed (step S55). Regardless of whether it is selected as the repack processing target, the write count counter of the corresponding logical stripe is cleared to 0 during the repack target stripe selection processing (step S55).

Thereafter, as n + 1, the logical stripe with the second largest number of writes (L1 = A [1] .Logic)
Pay attention to al_Stripe_ID) (step S5)
7, S58, S53), and the write count (A
[1]. It is determined whether Modified_Count) is equal to or less than the constant value D (step S54).

Write counter "A [n] .Mod
The process ends when a logical stripe in which "if_Count" is equal to or smaller than the constant value D appears (YES in step S54). Therefore, it is possible to give priority to a repacking target by giving priority to a logical stripe that has been written many times. In addition, the write counter “A []. Modified_Co
Since the "repack target stripe selection processing" is performed only for the logical stripes whose "unt" is equal to or greater than the constant value D,
The load required for the processing can be reduced.

In the second and third embodiments, the repacking target selection process based on α and β in the first embodiment should be preferentially performed in consideration of the read reference count value and the write count value. Although the logical stripe is determined, the logical stripe to be repacked is determined only by controlling the read reference count value and the write count value, or both the read reference count value and the write count value are used. Even when the logical stripe to be repacked is determined by the above method, a sufficient effect can be obtained.

As described above, the present embodiment is most effective when applied to a disk array having a RAID structure. However, in the case of a disk control system using the log structured write method, a disk device composed of one disk drive is used. The same can be applied to the control.

Further, by recording a computer program including the processing procedure of the RAID acceleration driver 100 in a computer-readable recording medium, the computer program can be implemented by simply introducing the computer program into the ordinary computer system through the recording medium. The same effect as the form can be obtained.

Further, the present invention is not limited to the above-described embodiment, and can be variously modified at the stage of implementation without departing from the spirit of the invention. Furthermore, the embodiments include inventions at various stages, and various inventions can be extracted by appropriately combining a plurality of disclosed constituent elements. For example, even if some constituent elements are deleted from all the constituent elements shown in the embodiment, the problem described in the section of the problem to be solved by the invention can be solved, and the effect described in the section of the effect of the invention can be solved. If you get
A configuration in which this component is deleted can be extracted as an invention.

[0071]

As described above, according to the present invention,
The data rearrangement process can be efficiently performed, and the read performance can be improved with a small load. In particular, by using the ratio of the number of valid blocks and the ratio of consecutive physical block numbers of blocks having adjacent logical block numbers, it is possible to efficiently select a logical stripe that needs rearrangement.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of a computer system according to an embodiment of the present invention.

FIG. 2 is a view for explaining the principle of write processing to a disk device provided in the computer system of the same embodiment.

FIG. 3 is a diagram showing an example of the contents of an address conversion table provided in the computer system of the same embodiment.

FIG. 4 is an exemplary flowchart showing a procedure of repack control processing executed in the computer system of the embodiment.

FIG. 5 is a diagram showing the configuration of a logical stripe management table provided in the computer system of the same embodiment.

FIG. 6 is a diagram showing how the contents of the address conversion table change before and after repack processing in the computer system of the same embodiment.

FIG. 7 is a diagram showing how the physical data arrangement changes before and after repack processing in the computer system of the same embodiment.

FIG. 8 is an exemplary flowchart showing a processing procedure at the time of data reading in the computer system of the same embodiment.

FIG. 9 is an exemplary flowchart showing a second procedure of repack control processing executed in the computer system of the embodiment.

FIG. 10 is an exemplary flowchart showing a processing procedure at the time of data writing in the computer system of the embodiment.

FIG. 11 is a flowchart showing a third procedure of repack control processing executed by the computer system of the same embodiment.

[Explanation of symbols]

11 ... CPU 12 ... Bridge 13 ... Main memory 16 ... RAID controller 17 ... Nonvolatile memory 18 ... Disk array 171 ... Log memory 172 ... Address conversion table 173 ... Logical stripe management table

Claims (14)

(57) [Claims] 1. A disk control system of a log structured write system, comprising: a disk array composed of a plurality of disk devices; a data buffer for storing a plurality of write-requested data blocks; and a storage in the data buffer. A data writing unit that generates write data for one physical stripe of the disk array from the plurality of created data blocks and collectively writes the write data to a predetermined physical stripe of the disk array, and a logical stripe of the disk array. An address conversion table that stores, for each logical stripe, the correspondence between a plurality of logical block numbers and a physical block number that indicates the physical position on the disk array where the data block specified by the logical block number exists. Referring to the address conversion table, The ratio of the number of valid logical blocks in the logical stripe and / or the ratio of consecutive adjacent physical block numbers is calculated, and if the result satisfies a predetermined condition, it is determined as the logical stripe to be relocated. Relocation target determining means, the data block of the logical stripe whose relocation is determined by the relocation target determining means is read from the disk array into the data buffer, and the data writing means reads an empty physical stripe area of the disk array. A disk control system, comprising: a data rearrangement unit for writing the data into the disk; and an address conversion table rewriting unit for rewriting the physical block number of the address conversion table to the physical block number of the free physical stripe. 2. The relocation target determining unit relocates a logical stripe in which a ratio of the valid logical blocks is larger than a predetermined value and a ratio of the adjacent physical block numbers is a predetermined value or less. The disk control system according to claim 1, wherein conditions are determined as an arrangement target. 3. A log-structured write-type disk control system, comprising: a disk array composed of a plurality of disk devices; a data buffer for storing a plurality of write-requested data blocks; and a storage in the data buffer. A data writing unit that generates write data for one physical stripe of the disk array from the plurality of data blocks that are written and collectively writes the write data to a predetermined physical stripe of the disk array, and a logical address space of the disk array. Multiple
Specified by the logical block number of
Object on the disk array where there is a data block
Store the correspondence with the physical block number that indicates the physical location
The address conversion table and the logical address space of the disk array are stored in one physical strut.
Type for each address range of the number of data blocks
For each logical stripe you cut, within that logical stripe
Data read times for multiple contained logical blocks
A counter unit for counting the number, a rearrangement target determining unit for determining, with reference to the counter unit, a logical stripe in which the number of times of data reading is a predetermined value or more as a logical stripe to be rearranged; Data relocation means for reading the data block of the logical stripe whose relocation is determined by the means from the disk array to the data buffer, and writing the data block in the free physical stripe area of the disk array by the data writing means; and the address translation table. And an address conversion table rewriting means for rewriting the physical block number of the physical block number of the free physical stripe. 4. A log-structured write-type disk control system, comprising: a disk array composed of a plurality of disk devices; a data buffer for storing a plurality of write-requested data blocks; and a storage in the data buffer. A data writing unit that generates write data for one physical stripe of the disk array from the plurality of data blocks that are written and collectively writes the write data to a predetermined physical stripe of the disk array, and a logical address space of the disk array. Multiple
Specified by the logical block number of
Object on the disk array where there is a data block
Store the correspondence with the physical block number that indicates the physical location
Address translation tables, one physical Stora logical address space of the disk array
Type for each address range of the number of data blocks
For each logical stripe you cut, within that logical stripe
Data write times for multiple contained logical blocks
A counter unit that counts the number, a relocation target determination unit that refers to the counter unit, and determines a logical stripe whose data write count is a predetermined value or more as a logical stripe that is a relocation target; and the relocation target determination unit. The data block of the logical stripe whose relocation has been determined by the data array is read from the disk array to the data buffer, and written in the free physical stripe area of the disk array by the data writing means; A disk control system comprising: an address conversion table rewriting unit that rewrites the physical block number to the physical block number of the free physical stripe. 5. The relocation target determining means refers to the address conversion table to calculate a ratio of the number of valid logical blocks in the logical stripe and / or a ratio of consecutive adjacent physical block numbers. 5. The disk control system according to claim 3, further comprising means for determining a logical stripe to be relocated if the result satisfies a predetermined condition. 6. A data rearrangement method for rearranging a data block stored in a disk array composed of a plurality of disk devices, wherein a plurality of write-requested data blocks are stored in a data buffer. Generating write data for one physical stripe of the disk array from the plurality of data blocks accumulated in the data buffer and writing the write data to a predetermined physical stripe of the disk array at once; For each logical stripe, an address conversion table showing a correspondence relationship between a plurality of logical block numbers forming a logical stripe of the array and a physical block number indicating a physical position on the disk array in which the data block of the logical block number is written is provided. And the step of generating By referring to the replacement table, the ratio of the number of valid logical blocks in the logical stripe and / or the ratio of consecutive adjacent physical block numbers is calculated, and when the result satisfies a predetermined condition, relocation is performed. The step of determining the target logical stripe, the data block of the logical stripe determined to be relocated is read from the disk array to the data buffer, and is written to the free physical stripe area of the disk array by the batch writing step to write data. A data relocation method comprising: a relocation step; and a step of rewriting the physical block number of the address conversion table to a physical block number of the empty physical stripe in association with the data relocation. . 7. A logical stripe in which the ratio of the valid logical blocks is larger than a predetermined value and the ratio in which the adjacent physical block numbers are continuous is a predetermined value or less is conditionally determined as a relocation target. The data rearrangement method according to claim 6, characterized in that 8. A data rearrangement method for rearranging a data block stored in a disk array configured by a plurality of disk devices, wherein a plurality of write-requested data blocks are stored in a data buffer. step a, from said stored plurality of data blocks in the data buffer to generate a physical stripe write data of the disk array, and writing collectively a predetermined physical stripe of the disk array, said disk Multiples that make up the logical address space of the array
Specified by the logical block number of
Object on the disk array where there is a data block
Store the correspondence with the physical block number that indicates the physical location
The step of generating an address conversion table and the logical address space of the disk array are stored in one physical stream.
Type for each address range of the number of data blocks
For each logical stripes cut, to its logical Stra in the type
Data read times for multiple contained logical blocks
Counting the number, determining a logical stripe in which the number of times of data read is a predetermined value or more as a logical stripe to be relocated, and relocating the data block of the logical stripe determined to be relocated from the disk array. Reading to a data buffer, writing to a free physical stripe area of the disk array by the batch writing step to relocate data, and rewriting the physical block number of the address conversion table to a physical block number of the free physical stripe And a data rearrangement method comprising: 9. A data relocation method for relocating a data block stored in a disk array composed of a plurality of disk devices, wherein a plurality of write-requested data blocks are stored in a data buffer. step a, from said stored plurality of data blocks in the data buffer to generate a physical stripe write data of the disk array, and writing collectively a predetermined physical stripe of the disk array, said disk Multiples that make up the logical address space of the array
Specified by the logical block number of
Object on the disk array where there is a data block
Store the correspondence with the physical block number that indicates the physical location
The step of generating an address conversion table and the logical address space of the disk array are stored in one physical stream.
Type for each address range of the number of data blocks
For each logical stripe you cut, within that logical stripe
Data write times for multiple contained logical blocks
Counting the number of data, the step of determining a logical stripe whose data write count is a predetermined value or more as a logical stripe to be relocated, and the data block of the logical stripe whose relocation is determined from the disk array. Reading to a buffer, writing to a free physical stripe area of the disk array by the batch writing step to relocate data, and rewriting the physical block number of the address translation table to the physical block number of the free physical stripe. A data rearrangement method comprising: 10. In the step of determining a logical stripe to be relocated, the address translation table is referred to,
The ratio of the number of valid logical blocks in the logical stripe and / or the ratio of consecutive adjacent physical block numbers is calculated, and if the result satisfies a predetermined condition, the logical stripe to be relocated is determined. The data rearrangement method according to claim 8 or 9, further comprising steps. 11. For each logical stripe, a logical stripe is included therein.
Counts the number of data reads for each logical block
Further comprising counter means for controlling the rearrangement target determining means for determining the number of data reads.
The redistribute is given priority to the logical stripes with a predetermined value or more.
Performs processing to determine the logical stripe to be placed
The disk control system according to claim 1, characterized in that
Mu. 12. The logical stripes are included in each logical stripe.
Count the number of data writes to the logic block
Further comprising counter means for controlling the rearrangement target determining means,
The redistribute is given priority to the logical stripes with a predetermined value or more.
Performs processing to determine the logical stripe to be placed
The disk control system according to claim 1, characterized in that
Mu. 13. A logical stripe is included in each logical stripe.
Counts the number of data reads for each logical block
The number of valid logical blocks and / or adjacent to each other.
The ratio of consecutive physical block numbers satisfies the specified condition.
The process of determining whether or not to perform
Priority is given to the logical stripes whose number is equal to or greater than the specified value
7. The data rearrangement method according to claim 6, wherein
Law. 14. The logical stripes are included in each logical stripe.
Count the number of data writes to the logic block
The number of valid logical blocks and / or adjacent to each other.
The ratio of consecutive physical block numbers satisfies the specified condition.
The process of determining whether or not to perform
Priority is given to the logical stripes whose number is equal to or greater than the specified value
7. The data rearrangement method according to claim 6, wherein
Law.