JPH07311659A - Cache incorporated disk array device - Google Patents

Abstract

PURPOSE:To improve the throughput by shortening the operation time of writing-back operation. CONSTITUTION:New data after a writing process by a host system 100 are stored in a cache memory part 2 and managed and when old data before the writing-back operation to a magnetic disk device are left on the cache memory part 2, the old data are also managed by a cache management table part 1. For example, a block A'5 on the cache memory part 2 is indicated with a block address 11 of a table a' and the old block C'7 corresponding to it is indicated with a block address 13 of the table a'. Before the block A'5 is written back, a D flag 10 is set to '1' and the presence of the old block C'7 is represented by setting a B flag 12 to '1'. Then a disk array control part 16 uses the old block C'7 to generate new parity when the block A'5 is written back.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a disk array device incorporating a cache memory.

[0002]

2. Description of the Related Art A disk array device is a file device incorporating a plurality of magnetic disk devices using the "RAID architecture" devised at the University of California, Berkeley. In such a disk array device, each data block is distributed and stored in n-1 of the n magnetic disk devices, and parity is stored in the remaining one. And, these magnetic disk devices,
The host system operates as if it were one unit, and the access operation time is shortened by the parallel operation of the magnetic disk devices. Even in such a disk array device, a cache memory is provided in order to further speed up the access operation. The write-back method is also used for this cache memory.

[0003]

However, the above-described conventional disk array device with a built-in cache has the following problems. That is, at the time of write back, new parity must be generated from the latest data on the cache memory, old data read from the magnetic disk device and old parity, and new data and new parity must be written to the magnetic disk device. In other words, when writing back,
Two read accesses to old data and old parity are required, and two write accesses to new data and new parity. In an application environment in which a large number of write accesses are generated from the host system at short intervals, it is necessary to process the access request from the host system while performing the write back operation inside the device. For this reason, there is a problem that the throughput inside the device is significantly reduced, and the apparent access time to the host system cannot be shortened as compared with the case where the caching is not performed.

[0004]

The present invention has been made in view of the above points and is characterized by the following points. A write-back type cache memory unit is provided for temporarily storing new data before being written to the magnetic disk device after the write processing by the host system is completed. It manages whether or not old data before the host system performs write processing exists in the cache memory unit, and when the old data exists, the address in the cache memory unit where the old data is stored. A cache management table unit for managing When writing back new data stored in the cache memory unit to the magnetic disk device, refer to the cache management table unit, and if it is determined that there is old data in the area to be written back, write the old data. A disk array control unit for generating a new parity of the corresponding area by using the disk array control unit is provided.

[0005]

In the RAID5 disk array device with a built-in cache, new data after write processing by the host system is stored and managed in the cache memory unit, and old data before being written back to the magnetic disk device is stored in the cache memory unit. If it remains, the old data is also managed by the cache management table section. As a result, when the new parity is generated, if the old data remains in the cache memory unit, it can be used, and it is not necessary to read the old data from the magnetic disk device. Therefore, it is possible to reduce the number of times the old data necessary for generating new parity at the time of write back is actually read from the built-in magnetic disk device, and thereby to improve the throughput of the disk array device.

[0006]

The present invention will be described in detail below with reference to the illustrated embodiments. FIG. 1 shows an example of a management system of a cache memory incorporated in a disk array device according to the present invention. The apparatus of FIG. 1 includes a cache management table unit 1, a cache memory unit 2, and a disk array control unit 16. The cache management table unit 1 is composed of a plurality of tables a'and b '. In each table, as in the conventional case, the V flag 8, the D flag 10, the logical address display unit 9, the LRU forward / backward link pointers 14 and 15,
Also, a block address 11 for storing the latest data is stored. In addition to these, the B flag 12 indicating that old data before writing (hereinafter referred to as “old data”) exists in the cache memory unit 2 and its block address 13 when old data exists Is stored.

Before describing the disk array device of the present invention, the details of the configuration of the conventional disk array device will be described to clarify the advantages of the present invention. Figure 2 shows the conventional R
1 shows a typical configuration of an AID5 disk array device. The illustrated disk array device comprises a disk array controller 38 and magnetic disk devices 39 to 43. The disk array controller 38 and each of the magnetic disk devices 39 to 43 are connected by a disk interface 44.
Through 48. The disk array device is connected to the host system by the host interface 49. Built-in magnetic disk device 3
Although the numbers of 9 to 43 are not specified in the RAID architecture, the most common five unit configuration shown in FIG. 2 will be described below as an example.

FIG. 3 shows an example of the arrangement of data blocks in this RAID5 disk array device having five units. Here, the data block is an integral multiple of the sector capacity on the built-in magnetic disk device (usually several KBytes to several tens of KBytes), and is a basic unit of data management on the disk array device. Normally, this size is set close to the unit most frequently accessed by the host system. The disk array device generally behaves as if it were one magnetic disk device to the host system, but as shown in FIG. The data blocks 50 to 53 that are continuous above and the parity block 54 generated by taking the exclusive OR of all the data in these blocks are actually arranged on separate internal magnetic disk devices. To be done. That is, in the case of a disk array device of n units, each data block and a parity block generated from it are n
It is distributed and arranged in the built-in magnetic disk unit of the stand.

Since one parity block is generated for every (n-1) block, the RA which can be seen from the host system is built even though n magnetic disk devices are incorporated.
The capacity of the ID5 disk array device is the capacity of one magnetic disk device × (n−1). However, by generating and storing the parity block, even if one built-in magnetic disk device fails, all the data on the failed magnetic disk device can be generated from the data on the remaining magnetic disk device. It is possible. Therefore, the reliability of the data of the entire device can be made much higher than that of the magnetic disk device alone. Further, in response to a read request from the host system, data can be accessed in parallel to a plurality of built-in magnetic disk devices, so that high throughput can be achieved.

On the other hand, in response to the write request from the host system, it is necessary to generate and write the parity data, so that the throughput is lowered. This RAID
Figure 5 shows the procedure of the write operation in a 5-disk array device.
Shown in. In FIG. 4, the disk array device includes a disk array controller 55, a built-in magnetic disk device 56 having a storage area 62 of a data block to be written, and a built-in magnetic disk having a storage area 63 of a parity block of the data block. It comprises a device 57. The write data transferred by the write request from the host system is
First, it is stored in the storage unit 58 on the disk array control unit 55. Furthermore, the data block before being written is read from the area 62 on the magnetic disk device 56, and the storage unit 5
9, the parity block is read from the area 63 on the magnetic disk device 57 and stored in the storage unit 60. Then, the exclusive OR of all the data in the storage units 58, 59, 60 is stored in the storage unit 61. The data on the storage unit 61 thus generated is a new parity block.

That is, the data block before writing is di,
The parity block is dp, and the remaining data blocks that are the source of this parity block are dj and d, respectively.
Assuming k, dl, and dm, the following relationships are established before writing. dp = di EXOR dj EXOR dk EXOR dl EXOR dm (Equation 1) Then, assuming that the write data is di ′ and the new parity data is dp ′, the following relations must be satisfied. dp '= di' EXOR dj EXOR dk EXOR dl EXOR dm (Equation 2) Here, since d EXOR d = 0 and 0 EXOR d = d,
When the right side of Equation 2 is transformed,

Dp '= (di' EXOR dj EXOR dk EXOR dl EXOR dm) EXOR 0 = di 'EXOR dj EXOR kk EXOR dl EXOR dm EXOR di EXOR di = (di EXOR dj EXOR kk EXOR dl EXOR dm EXOR dm EXOR dm EXOR dm EXOR dm EXOR dm R di ′ = dp EXOR di EXOR di ′ (Equation 3) This makes it possible to prove that the data generated on the storage unit 61 is a new parity block according to the procedure shown in FIG. After generating new parity data in this way,
By writing the write data of the storage unit 58 to the area 62 on the magnetic disk device 56 and the new parity data of the storage unit 61 to the area 63 on the magnetic disk device 57, the actual write operation as the RAID5 disk array device can be performed. Complete.

If the write request from the host system spans a sufficiently long area and new parity data can be generated from the write data itself by (Equation 1), it is not always necessary to read the data before writing from the magnetic disk device. Absent. However, since the block size of the RAID5 disk array device is often set in accordance with the unit most frequently accessed by the host system, such a case can be said to be rare. As described above, in the RAID5 disk array device, it is usually necessary to perform two read accesses and two write accesses inside the device in response to one write request from the host system. Compared with this, there is a problem that the throughput inside the apparatus is extremely reduced.

In order to avoid this problem, a technique of incorporating a cache memory in the disk array controller is often used. FIG. 5 shows an example of the configuration of a control unit of a disk array device having a built-in cache memory and its write operation. In the device of FIG. 5, a cache memory unit 66 and a cache management table unit 65 are provided in the disk array control unit 64. The cache memory unit 66 normally uses the size of a data block on the disk array device as a basic unit, and is managed by the cache management table unit 65 for each cache block formed by dividing the cache memory unit 66 into the size.

FIG. 5A shows an operation when the data of the area to be written exists in the cache memory unit 66 (hereinafter, this is called a write hit). In this case, the write data 68 from the host system is stored in the corresponding area 67 on the cache memory unit 66, and the write completion is notified to the host system when this is completed. In this way, the write end is notified before actually writing the write data and the parity data corresponding thereto to the magnetic disk device, so that the apparent access time can be greatly shortened.

FIG. 5B shows an example of the operation when the data to be written does not exist in the cache memory unit 66 (hereinafter referred to as a write miss). In this case, 'the data is read from the magnetic disk device 69 in which the data of the corresponding area is stored and stored in the allocated area 71 on the cache memory unit 66, and then the'write data 70 is stored in the area 71. Here, the data in the write target area is read into the cache memory unit 66 because the management of the cache memory unit 66 is normally managed in block units, but the unit of write access from the host system does not always match this. This is because there is no valid data and it is necessary to store valid data in the remaining area on the block that is not the write target.

Of course, if the area of the write access from the host system coincides with the block on the cache memory unit 66, the write data need only be written directly on the cache memory unit 66. Access time can be achieved. Further, when managing the validity / invalidity of data for each sector in a cache block, at the time of a write miss, write data is stored in the allocated cache block and only the area written in that block is registered as a valid sector. Good. By doing so, although the cache management becomes a little complicated, the read operation of the data before the write can be made unnecessary.

FIG. 6 shows an example of the management system of the cache memory unit. As shown in this figure, the cache management table 72 allows the cache memory unit 73 to
Is managed. The cache memory unit 73 is managed by being divided into blocks, and tables a and b on the cache management table 72 exist for each valid block. This table includes a flag (hereinafter referred to as a V flag) 78 indicating validity / invalidity of the table itself, a logical address display unit 79 indicating a logical address of data stored in a corresponding cache block, and a corresponding cache block. A flag (hereinafter referred to as a D flag) 80 which is set when the write data is stored in and is reset when the writing to the magnetic disk device is actually completed, a block address display section 81 which indicates the address of the corresponding cache block, and The blocks are accessed most recently (hereinafter referred to as LRU order. LRU is Le
It is composed of a forward link pointer 82 and a backward link pointer 83 for linking to "ast recently used".

In FIG. 6, the cache memory unit 7 is used as an example.
2 blocks on block 3, block A76 and block B7
7 and table a74 and table b75 on the cache management table corresponding to each of them. In FIG. 6, since the D flag of the table a74 is 1,
It can be seen that block A76 holds write data from the host system that has not yet been written to the built-in magnetic disk device. On the other hand, the data on the block B77 is a copy of the data on the magnetic disk device because the D flag of the table b75 is 0. Then, the table a74 and the table b7
In V, the V flag is 1 because a valid data block is displayed.

A disk array device incorporating such a cache memory normally refers to the cache management table 72 on a regular basis, and if there is a block whose D flag is "1", it has not been accessed most recently. That is, the lowest one of the LRU links is selected in order, and the data is written to the corresponding area of the magnetic disk device corresponding to the logical address on the table (this is called write back). After a sufficient time elapses in this way, all write data from the host system is reflected in the built-in magnetic disk device.

Next, a disk array device with a cache according to the present invention will be described. As shown in FIG. 1, according to the present invention, the basic configuration of the entire device and the arrangement of data blocks are the same as those of the conventional device shown in FIGS. 2 and 3, respectively.
The only difference is that the fields of the B flag 12 and the block address 13 of the old data are added to each table of the cache management table unit 1. In FIG. 1, since the D flag is 1 in the table a'3, the block A '
5 stores write data from the host system that has not been written back, and the B flag is set to 1
Therefore, the old data before writing is block C'7.
You can see that it is stored in. Also, the table b'4
Since the D flag is 0, it can be seen that the data stored in the block B'6 is a copy of the data on the built-in magnetic disk device.

FIG. 7 shows an example of the configuration of the control unit of the disk array device with a built-in cache according to the present invention and its write operation. In FIG. 7, the disk array device comprises a disk array control unit 16, a cache management table unit 17, and a cache memory unit 18. Figure 7
(A) shows the operation at the time of a write hit. In the case of FIG. 7A, the write data 21 from the host system is stored in the newly allocated block 19 on the cache memory unit 18. If the entire block cannot be filled with the write data 21 from the host system, the remaining part of the data is copied to the block 19 on the cache memory from the block 20 in which the data before writing is stored. Then, when these operations are completed, the write completion is notified to the host system.

Since both the write data and the old data are managed in the cache memory unit in block units, it may be necessary to copy the data in the cache memory unit at the time of a write hit. However, since the time required to copy data on the cache memory unit is much shorter than the time required to access the magnetic disk device, the increase in access time is small compared to the conventional method. Further, when managing the validity / invalidity of data for each sector in the cache block, if only the area storing the write data is registered as a valid sector, this data copy operation becomes unnecessary and the access time is the same as before. It is possible to achieve.

FIG. 7B shows an example of the operation at the time of a write miss. In this case, 'the data is read from the magnetic disk device 22 in which the data of the corresponding area is stored and allocated on the cache memory unit 18. It is stored in the block 25, and the'write data 23 is stored in the separately allocated block 24. Further, if necessary, the data is copied from the block 25 storing the data before the write to the block 24 on the cache memory portion to the remaining area where the valid data is not stored. Notify the host system of the end of writing. Also at this time, if the validity / invalidity of data is managed for each sector in the cache block, the data copy operation on the cache memory unit can be omitted as in the case of the write hit. In that case, it is not always necessary to read the data before writing from the magnetic disk device at this time, and the data before writing may be read for the first time at the time of performing write back.
In either method, the time required for copying the data on the cache memory unit is increased at most and the access time is slightly increased compared with the conventional method.

8 and 9 show how the cache management table is updated at the time of a write hit. FIG. 8 shows a state before update, and an area indicated by the logical address display unit 9 by referring to the table a12 (the V flag 8 is 1 and this table is valid) on the cache management table unit 1. Of the block A27 in the cache memory unit 2 having the address indicated by the block address 11
You can see that it is stored in. Further, since the D flag is 0, it can be seen that this data on the cache memory unit is a copy of the data on the built-in magnetic disk device. At this time, the D flag 10 is 0, so the B flag 12 is "0", and the block address 13 pointing to the storage block of the old data is invalid.

FIG. 9 shows a state after the area on the block A27 is write-hit and the write data is stored in the newly allocated block B28. Due to the write hit, the data on the block A27 is the old data before the write. Therefore, the block address is set to the block address 13 indicating the old data storage block of the table a26, and the B flag 12 is set to "1". Further, the address of the block B28 is set to the block address 11 indicating the storage block of the latest data, and the D flag 10 is set to "1". By operating the cache management table as described above, it is possible to manage the data after writing and the old data before writing in the cache memory unit.

Next, FIG. 10 shows the operation at the time of write back when both data are present in the cache memory unit.
Shown in. In FIG. 10, the disk array device includes a disk array controller 29, a built-in magnetic disk device 30 having a storage area 36 of a data block to be written back, and a built-in magnetic disk device having a storage area 37 of a parity block of the data block. It is composed of a disk device 31. The write data transferred by the write request from the host system is the block 3 of the cache memory unit on the control unit 29.
Stored in 2. Then, it is assumed that a copy of the data (old data) in the area 36 on the magnetic disk device 30 is already stored in the block 33 of the cache memory unit.

Therefore, first, the parity block is read from the area 37 on the magnetic disk device 31 and stored in the allocated area 34 on the cache memory unit. Then, the blocks 32 and 3 on the cache memory unit are next.
The new parity generated by taking the exclusive OR of all the data of 3 and 34 is stored in the cache block 35. After the new parity is generated in this way, the write data of the block 32 is written to the area 36 on the magnetic disk device 30, and the new parity of the block 35 is written to the area 37 on the magnetic disk device 31. This gives R
The actual write operation as the AID5 disk array device is completed.

As described above, according to the method of the present invention, when old data before writing exists in the cache memory unit, the read access is performed once for accessing the built-in magnetic disk device required for write back. 2 times, the actual access to the magnetic disk device can be reduced by one time compared to the conventional method, and the throughput inside the device can be improved. In particular, when normal read data caching is performed, in which the data in the corresponding area on the magnetic disk device is loaded into the cache memory unit at the time of a read miss, as well as the caching of write data, when previously read and loaded into the cache memory unit. If a write hit occurs in an area that has been written, or if a write occurs again in an area in which the data in the cache memory unit has already been written back and the data in the cache memory unit matches the data in the magnetic disk device, the throughput is guaranteed. The improvement effect can be exhibited.

Therefore, the effect is great when a write access frequently occurs in the same area after a read access or when a large number of write accesses occur in a relatively narrow area. And this situation is R
In the online transaction processing environment in which the AID5 disk array device is suitable, the frequency of occurrence is relatively high, and a sufficient effect can be expected. This is because the number of accesses per unit time is very large and the frequency of write access is particularly high in online transaction processing. Furthermore, most database management systems used for online transaction processing manage their own cache memory on the main storage device. Therefore, when updating data, it is possible to write once after reading the area containing the data. This is because it is necessary to write data to a log file having a relatively small capacity for each update in order to keep an update log of data.

11 and 12 show how the cache management table is updated after write back. FIG. 11 shows the state of the cache management table before write back.
This is equivalent to the state after the write hit in FIG. And FIG.
Shows the state after the write back. Since the write-back is completed, the old data before the write is no longer necessary and the block A27 is invalidated and the area is released. Then, the D flag 10 and the B flag 12 are reset to the original "0", and the block address 13 where the old data is stored is stored.
Is invalid. In this way, after the write back, the same state as in FIG. 8 in which a copy of the data on the built-in magnetic disk device is stored is entered.

The present invention is not limited to the above-mentioned embodiments, and it goes without saying that various modifications can be made. For example, in the embodiment described above, RAID
Although the case of a 5-level disk array device has been described, the present invention is not limited to this and can be applied to a RAID 4-level disk array device.

[0033]

As described above in detail, according to the present invention, in a disk array device having a cache, a cache management system capable of holding both write data from the host system and old data before writing in the cache memory thereof. In addition, when writing back, the old data before writing in the cache memory is used when generating new parity. This makes it possible to reduce the number of times of actual access to the built-in magnetic disk device at the time of write back by one. Therefore, even if the access request from the host system is processed while the write-back operation is performed inside the device in the application environment where a large number of write accesses are generated from the host system at short intervals, compared to the conventional method. The rate of decrease in throughput inside the device can be reduced. Therefore, the apparent access time to the host system can be shortened as compared with the conventional case.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of a disk array device with a cache according to the present invention.

FIG. 2 is a block diagram showing a typical configuration of a RAID 5 disk array device.

FIG. 3 is an explanatory diagram of an arrangement example of data blocks in a RAID 5 disk array device.

FIG. 4 is an explanatory diagram of a write operation in the RAID5 disk array device.

FIG. 5 is an explanatory diagram of a configuration of a control unit of the disk array device with a cache and its write operation.

FIG. 6 is an explanatory diagram of a conventional cache memory unit management system.

FIG. 7 is an explanatory diagram of a configuration of a control unit of a disk array device with a cache according to the present invention and its write operation.

FIG. 8 is an explanatory diagram of a state of a cache management table before writing.

FIG. 9 is an explanatory diagram of a state of a cache management table after writing.

FIG. 10 is an explanatory diagram of the write-back operation of the disk array device with cache according to the present invention.

FIG. 11 is an explanatory diagram of a state of a cache management table before write back.

FIG. 12 is an explanatory diagram of a state of the cache management table after write back.

[Explanation of symbols]

 1 cache management table unit 2 cache memory unit 16 disk array control unit 100 host system

Claims (1)

[Claims] 1. A write-back type cache memory unit for temporarily storing new data before being written to a magnetic disk device after a write process by the host system is completed, and the host system writes on the cache memory unit. A cache management table unit that manages whether or not old data before processing exists and manages an address on the cache memory unit where the old data is stored when the old data exists, When new data stored in the cache memory unit is written back to the magnetic disk device, the cache management table unit is referred to, and if it is determined that there is old data in the area to be written back, the old data is deleted. A disk that uses the data to generate new parity for the corresponding area. Cache internal disk array device characterized by consisting of an array controller.