JPH07200187A - Disk array device - Google Patents

Abstract

PURPOSE:To provide a disk array device which can carry out fast input/output operation, facilitates disk management, and causes no decrease in reliability even when many disk drives are connected. CONSTITUTION:This disk array device is equipped with a global disk array controller 103 which puts local disk array controllers 104 to 107, where plural disk drivers 137 to 140 are connected, further in a disk array, Each disk array controller performs RAID control over plural disk drives in its machine. The global disk array controller performs RAID control over the local disk array controllers. Consequently, highly reliable disk devices are obtained by fastness by the division of data and global parity which are generated by the local disk arrays and global disk array and stored in the local disk arrays.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a disk file system used in a computer system or the like, and more particularly to a high speed and highly reliable disk array device.

[0002]

2. Description of the Related Art Generally, a computer system includes a processor and a secondary storage device. A magnetic disk device is a secondary storage device that is often used. In the magnetic disk device, the capacity growth rate is extremely high at present, but the performance of the magnetic disk apparatus accompanied by mechanical operation is not as high as the processor performance growth rate. A disk array has been proposed as a method for solving the problem.

As a representative paper on the disk array, "A of D. Patterson, G. Gibson, and RH Kartz et al.
Case for Redundant Arrays of Inexpensive Disks (RAI
D), in ACM SIGMOD Conference, Chicago, IL, (June 198
8) ” RAID is a technology that shortens access time by allocating data to multiple disk drives in a distributed manner, and also improves reliability by storing redundant data called parity for error correction or ECC. Is. In other words, the high speed that can be input and output to multiple disk drives in parallel,
This is a technology that can recover the data of the failed disk drive from the data other than the parity and the failed disk drive even when a failure occurs in any of the disk drives.

A patent application for realizing this technique has also been filed, for example, Japanese Patent Laid-Open No. 4-230512 by Jaishyanker Musedas Minon and James Matthews Carson (International Business Machines CORP). This publication discloses a technique for speeding up parity update, which is one of the problems in the disk array.

In the disk array, the data itself and the parity must be updated when writing the data. In order to update the parity, there are cases in which the data and parity before the update are required. Therefore, the overhead of reading these data and parity in advance and the overhead of writing new data and parity at the original data and parity positions. Takes. This is called the light penalty. The above-mentioned publication discloses a technique for reducing the overhead due to the write penalty by storing new data and parity in the free area of the disk drive without recording them in the original position.

Another technique for facilitating the mounting of the disk array is disclosed in Japanese Patent Laid-Open No. 5-46524. The disk array device described in this publication connects a plurality of disks to a common bus to facilitate mounting of the disk array, and a host controller connected to the bus controls the disk array.

[0007]

The basic functions of the disk array controller are to interpret / execute input / output requests from the host processor, divide data, generate parity, manage cache memory, and start / end disk drive. There is processing. As described above, the disk array controller has to perform more processing than the normal disk controller.

For example, assume that a RAID 5 disk array device is configured with two drives (one disk drive for storing data and one disk drive for storing parity). At this time, when writing new data, the old data and the old parity must be read from each of the two disk drives, and then the new data and the new parity must be output to the two disk drives. In other words, one output process results in four disk inputs / outputs, which is a write penalty. In addition, it is necessary to perform data division, parity generation, etc. as processing unique to the disk array controller.

In the conventional disk array device, one disk array controller controls a plurality of disk drives. In order to control a large number of disk drives with such a configuration, it is necessary to provide a high speed disk array controller or a plurality of similar disk array controllers.

However, a high speed disk array controller is very expensive, which leads to an increase in the cost of the entire disk array device. Further, if a plurality of similar disk array controllers are provided, the user sees them as a plurality of disk devices, which has the drawback of complicating disk management.

An object of the present invention is to improve a disk array device. Further, an object of the present invention is to enable a high-speed input / output even if a large number of disk drives are connected, which does not increase the cost, and which makes disk management easy and does not reduce reliability. To provide a device.

[0012]

According to the present invention, a plurality of disk drives are divided into a plurality of logical groups each including a plurality of disk drives, and each logical group is provided for each logical group, for example. Each is managed by the disk array controller. Then, based on the address in the volume and the data length transferred from the host processor, one or more of the logical groups are selected, a parity is generated from the data stored in each of the selected logical groups, and the logical The data and parity are stored in the disk drives in the group.

Further, to the other disk drives other than the drive in which the parity is stored in the logical group, the consecutive addresses in the volume are sequentially assigned for each storage unit of the disk drive, and the addresses in the volume are plural. Allocate horizontally to the logical groups sequentially.

[0014]

The present invention is divided into a plurality of logical groups composed of a plurality of disk drives, and for example, a disk array controller is provided for each logical group for management, so that an inexpensive disk array controller for managing a small number of disk drives, Further, to the other disk drives other than the drive in which the parity is stored in the logical group, the consecutive addresses in the volume are sequentially assigned for each storage unit of the disk drive, and further, the addresses in the volume are plural logical groups. By horizontally allocating the data sequentially over, the user can manage as a single disk drive.

[0015]

Embodiments of the present invention will be described in detail below with reference to the drawings.

[Embodiment 1] FIG. 1 is a diagram showing the overall configuration of a disk array device according to a first embodiment of the present invention.

The disk array controller (102) is connected to the host processor (101). The disk array control device (102) includes a plurality of disk devices (1
33,134,135,136). Each disk device (133,134,135,136)
Multiple disk drives (137, 138, 1)
39, 140).

The disk array controller (102) comprises an interface controller (108), a global disk array controller (103), and a plurality of local disk array controllers (104, 105, 106, 107). ing.

The interface control unit (108) controls the transfer protocol of data or commands with the host (101). The global disk array control device (103) divides data, generates / stores parity, or controls a plurality of local disk array control devices for the plurality of local disk array control devices (104, 105, 106, 107). (104,10
5, 106, 107) and merge processing of the data transferred.

One local disk array controller (104, 105, 106, 107) corresponds to one logical group. Each local disk array control device (104, 105, 106, 107) performs divided storage of data received from the global disk array control device (103) and generation / storage of parity. A logical group is a unit of RAID control. That is, the local disk array control devices that make up the logical group are
The data received from the host is divided, parity is generated and stored in a plurality of drives, and the unit of RAID control including the plurality of drives is called a logical group.

Global Disk Array Controller (10
3) and a plurality of local disk array controllers (10
4, 105, 106, 107) both cause a failure of the disk device (133, 134, 135, 136) and a disk drive (137, 138, 139, 140) in the disk device (133, 134, 135, 136). In the event of a failure, parity can be used to recover from the failure.

One of the characteristics of this embodiment is that the local disk array controller (104, 105, 106, 107)
For controlling the global disk array controller (10
3) is provided. These related detailed operations will be described later.

Global disk array controller (10
3) is a command controller (141) and a control memory (14
5), cache memory (142), global RAI
It is composed of a D control unit (143) and an interface controller (144).

The command control unit (141) interprets an input / output request from the host processor (101) and queues the input / output request in a queue in the control memory (145). Control information for managing the entire disk array device is stored in the control memory (145). The data input / output by the disk array device (102) is stored in the cache memory (142) within the capacity of the cache memory (142). Data that is frequently accessed has a high probability of remaining in the cache memory (142).

The global RAID controller (143) is
A plurality of local disk array control devices (104, 10
5, 106, 107) is controlled by the host 101 so that it can be handled as a single disk or a plurality of disks. Interface controller (14
4) is a local disk array controller (104, 1)
05, 106, 107) and transfer control of data or commands.

The above is the outline of the configuration of the disk array device of this embodiment, and a detailed description will be given below.

FIG. 2 shows an example of the outline of the operation of the present invention, and an example of the features and effects of the present invention will be described below.

The outline of this operation is as follows.
Data (223) from the disk drive (133, 1)
34, 135) and the flow of data until the data is stored in the disk drive. In this example, the local disk array controller (104) performs the RAID3 operation.

The data (223) transferred from the host processor (101) is transferred to the disk array controller (10).
2) Global disk array controller (103)
Then, the local disk array controller (104, 10
(5,106) divided into three (3) (20
1). These data (219, 220, 221) are
Parallel local disk array control devices (104, 10
5, 106).

Local disk array controller (10
Data transfer to the local disk array controller (104, 105, 106)
Within each local disk array controller (104, 1
05, 106), the data is divided into the number of disk drives connected to the number of disk drives connected to (05, 106) (207, 209, 211), and at the same time, the global disk array controller (10
Parity of the data transferred from (3) is generated (2
08, 210, 212). The data and the parity divided in the local disk array controller (104, 105, 106) are stored in parallel in the disk drive.

Local disk array controller (10
4, 105, 106) each have a mechanism capable of operating independently, so that all disk drives can operate in parallel. With such a configuration, high-speed input / output of large-scale data becomes possible.

FIG. 3 shows an example of the outline of the operation of the present invention, and an example of the features and effects of the present invention will be described below.

The outline of this operation is as follows.
The operation and the flow of data from 1) to the storage of data (223) in the disk drive in the disk device (133) are shown. In this example, the local disk array controller (104) performs the RAID5 operation.

When the data (223) is transferred from the host processor (101), the global disk array controller (1) in the disk array controller (102).
In 03), the local disk array controller (104) to store the data is selected.

Local Disk Array Controller (10
When the data is transferred to 4), the number of disk drives connected to the local disk array controller (104) minus the number of disk drives for parity storage is subtracted from the number of disk drives connected to the local disk array controller (104). The data is divided (207), and at the same time, the parity of the data transferred from the global disk array controller (103) is collectively generated (208).

Since the normal RAID 5 performs input / output in block units, a write penalty occurs in each block when storing a plurality of blocks. However, as shown in this example, when a plurality of blocks are continuously written, as in the present invention, by continuously setting the addresses,
It becomes possible to collectively generate parity.

Local disk array controller (10
The data and parity divided in 4) are stored in parallel in the disk drive. With such a configuration, the throughput of short data can be improved.

FIG. 4 shows an example of the outline of the operation of the present invention, and an example of the features and effects of the present invention will be described below.

The outline of this operation is as follows.
Data (223) from the disk drive (133, 1)
Disk drives (13, 34, 135, 136)
7, 138, 139, 140) and the data flow.

The data (223) transferred from the host processor (101) is transferred to the disk array controller (10).
2) Global disk array controller (103)
Then, the local disk array controller (104, 10
5, 106, 107) and the number of local disk array control devices (107) for storing global parity is removed (201). At the same time, global parity of the entire data transferred from the host processor (101) is generated (202). These data are sent in parallel to the local disk array controller (10
4, 105, 106, 107).

In general, a disk array generates a parity different from data and stores it in a disk drive in order to improve reliability. Even if a failure occurs in the disk drive, the data of the failed disk drive can be calculated from the parity and the data of the disk drives other than the failed disk drive.

Local disk array controller (10
4, 105, 106, 107), the local disk array controller (104, 10)
5, 106, 107), the disk drives (137, 138, 139, 1) connected to the local disk array controller (104, 105, 106, 107).
40), the data is divided into the number of units excluding the parity storage disk drive (207, 209,
211, 213). At the same time, the parity of the data transferred from the global disk array controller (103) is generated (208, 210, 212, 21).
4). Local disk array controller (104, 10
5, 106, 107) and the data and the parity are divided into disk drives (137, 138, 13) in parallel.
9, 140).

Local Disk Array Controller (10
4, 105, 106, 107) have a mechanism capable of operating independently, so that all the disk drives (137, 138, 139, 140) can operate in parallel. With such a configuration, high-speed input / output of large-scale data becomes possible.

In addition to the normal disk array parity, the local disk array controller (104,
(Parity between 105 and 106) (global parity)
High reliability can be obtained from the fact that is generated / stored. In other words, recovery processing in each local disk array control device (104, 105, 106, 107),
This is because the failure recovery processing is multi-layered in the present invention, such as the recovery processing in the higher-level global disk array controller (103).

FIG. 5 shows an example of features and effects different from the features and effects described with reference to FIGS. 2 to 4. In particular,
7 shows a flow of an input / output request and a termination report according to the present invention.

When data is transferred from the host processor (101), the global disk array controller (1
03) divides the data and generates global parity, and the lower local disk array controller (10)
Input / output is performed in parallel with (4), 105, 106, 107) (301). Local disk array controller (10
4, 105, 106, 107) divides the data transferred from the global disk array controller (103) and generates parity, and the disk drive (137, 1)
Input / output (30, 139, 140) in parallel (30
6, 307, 308, 309).

Now, assuming that the number of local disk array control devices (104, 105, 106, 107) is m, each local disk array control device (104, 105, 1).
06, 107) and the number of drives connected is d,
There are a total of md disk drives. If all the md disk drives are connected to one disk array control device, one control device will handle the input / output of md times, and a very high speed control device is required. .

However, in the present invention, it is only necessary to process a small number of times of input / output, d times, for each local disk array control device (104, 105, 106, 107), and the local disk array control device (104, 105, 10
6, 107) does not require high-speed processing.

FIG. 6 shows the structure of the command control unit (141). The command control unit (141) includes an adapter (401), a processor (402), which controls transfer of commands and data with the interface control unit (108),
DMAC (403), memory (404), bus (40
7) and a buffer (408).

Microprograms (405, 406) for command control are stored in the memory (404). Specifically, the command reception processing program (4
05) and the end processing program (406). These details will be described later.

Micro Program (405, 406)
Via the common bus (408) to the processor (40
It is executed in 2). Data transfer between the command control unit (141) and the external control unit is executed by the DMAC (403) which is a direct memory access controller, independently of the processor (402). Command controller (1
Data and commands are transferred from 41) to the control memory (145) or cache via the buffer (408) and the bus (409).

FIG. 7 shows the tables and lists stored in the control memory (145). 502 is a virtual drive management table, 503 is an I / O management list, 50
Reference numeral 4 is a local disk array management table, 505 is a cache management list, and 506 is a command queue list.

The virtual drive management table (502) is
A plurality of local disk array control devices (104, 10
5, 106, 107), and controls the area set according to the user's request. The I / O management list (503) is a table that manages the input / output requests received from the host processor (101) together with the input / output states and the like until the input / output becomes the environment.

Local disk array management table (5
04) is the global disk array controller (10
This is a table for managing the local disk array control devices (104, 105, 106, 107) connected to 3), and stores the RAID level, capacity, partition configuration and the like. Cache management list (5
In 05), the data stored in the cache memory (142) and its address are stored. The command queue list (506) stores management information for the global disk array controller (103) to add tags to commands.

The detailed structure of these tables and lists pointed to by the pointer 501 of the control memory 145 will be described later.

FIG. 8 shows the structure of the local disk array management table (504).

The column 601 stores the identifier of the local disk array. The column 602 stores the partition identifier of the local disk array. With this column, a single local disk array area can be divided into a plurality of sections. This is because the disk array is composed of multiple disk drives and may be too large to be used as a single area. This column can be arbitrarily designated by the user.

A column 603 stores a path identifier. The global disk array controller (103) is the local disk array controller (104, 105, 1).
06, 107) when issuing an I / O request
The position of the local disk array controller (104, 105, 106, 107) can be known by referring to this path identifier. A column 604 is for each local disk array control device (104, 105, 106, 10).
It shows the number of drives owned in 7). It is used when calculating the parity position when the global disk array controller (103) creates global parity.

A column 605 shows the RAID level controlled by each local disk array controller (104, 105, 106, 107). This is also column 6
Similar to 04, it is used when calculating the parity position when creating global parity. In the column 606, each local disk array controller (10
4, 105, 106, 107) stores the number of disks storing parity. Like column 605, this is also used when calculating the parity position when creating global parity.

The column 607 stores the striping size for each partition of each local disk array controller (104, 105, 106, 107).
The striping size means the minimum input / output unit of the local disk array controller (104, 105, 106, 107). For example, the striping size is 4
In (KB), data having a data length of 4 KB or less is stored in a single disk drive. Data having a length exceeding 4 KB is divided and stored in the next disk drive every 4 KB.

The column 608 stores the capacity of each partition. The column 609 stores the status, for example, the status of each local disk array control device (104, 105, 106, 107) such as occurrence of a failure.

FIG. 9 shows the virtual drive management table (50
The configuration of 2) is shown. The virtual drive management table is provided for mapping the disk drive recognized by the host processor (101) and the physical disk drive.

The column 701 stores virtual drive identifiers. In columns 702 to 705, the virtual drive indicated by the virtual drive identifier is the local disk array controller (104, 105, 106, 10).
It shows which partition of 7) is configured. For example, the virtual drive VOL1 is a local disk array control device (104, 105, 106, 107).
Partition 1 of the above.

The column 706 stores the virtual drive capacity. The column 707 stores the global RAID level. This global RAID level corresponds to the local disk array controller (104, 10).
5, 106, 107) can be set regardless of the RAID level. The column 708 stores the number of disks storing parity. The striping size of the global disk array controller (103) is stored in the column 709. A column 710 stores a status, for example, virtual drive management information such as a failure occurrence.

FIG. 10 shows how the virtual drive is viewed from the host processor (101) and how data is physically arranged by using the virtual drive management table (502) shown in FIG. It is a conceptual diagram which shows whether or not.

Input / output requests (801, 802, 803) are issued from the host processor (101). At this time, the host processor (101) uses the virtual drive (8
04, 805, 806). For input / output to / from such a virtual drive, physically, the disk device (133, 134, 135, 1
The data is distributed in 36). Virtual drives (804, 805, visible from the host processor (101)
806) is physically LDRV1 (807), LD
Like the RV2 (808) and LDRV3 (809), they are distributed across the disk devices (133, 134, 135, 136).

By changing the contents of the virtual drive management table (502), the disk device (133,
(134, 135, 136) is not distributed over the entire structure. It is also possible to define one local disk array area as one virtual drive.
Further, 810, 811, 812 are parities generated / stored by the local disk array controller (104, 105, 106, 107), and 813 is parity generated / stored by the global disk array controller (103). ing. With such a table, the capacity required by the user can be satisfied, and the RAID level and the like can be freely set, so that there is flexibility in performance.

FIG. 11 shows the structure of the I / O management list (503).

The address pointed to by the control memory (145) indicates two more pointers.
One is a pointer (901) to the area where the list of the next input / output requests is stored, and the other is a pointer (902) that indicates the area where the input / output request is stored.
Reference numeral 909 is an identifier indicating the end of the list.

The stored input / output requests are shown in the range 903 to 908. Reference numeral 903 stores a host identifier for identifying from which host processor the request is made, 904 stores the time when the input / output is received, and 905 stores the input / output state. The input / output state is, for example, that input / output is being executed. A command 906 stores the command transferred from the host processor.

907 stores a tag added when a command is transferred from the host. This tag is for the host processor that issued the I / O to identify which I / O request was completed. This is added when the host processor can issue multiple I / O requests to the same device. If this tag is not present, the host processor cannot identify which input / output is completed even if the end notification is returned from the same device.

908 stores a new tag. This new tag (908) is set by the global disk array controller (103), and the tag (90
7) is for the host to identify the input / output, the new tag (908) is for the global disk array controller (10).
3) is the same local disk array controller (104,
(105, 106, 107) and is used when a plurality of input / output requests are issued.

The reason why the new tag (908) is necessary is that the virtual drive identified by the host and the local disk array control device (104, 105, 106, 107) identified by the global disk array control device (103) are used. Because they are not the same. That is, the tag adding side adds a unique tag for each drive. However, if this device is used by a device that receives the tag for a drive different from the drive recognized by the host, the uniqueness of the tag for each drive will be lost. Therefore, as the tag used by the global disk array controller (103), it is necessary to use the new tag (908) which can be managed by the global disk array controller (103), not the tag transferred from the host. is there.

FIG. 12 shows the cache management list (50
The structure of 5) is shown. Cache management list (50
5) is the location of the data cached in the cache memory (142) and the disk drive (137,
138, 139, 140) and the like.

The address pointed to by the control memory (145) indicates two more pointers.
One is a cache memory free pointer (1001)
And the other is the in-use pointer (1002). The cache memory (142) is managed in block units, and each of the lists (10) pointed to by the free pointer (1001) and the in-use pointer (1002).
02-1006, 1007-1011) corresponds to a fixed block.

The list (1002-1006) pointed to by the free pointer (1001) is used to manage the unused area (block) of the cache memory (142). If a new area of the cache memory (142) is to be acquired, the list (10) pointed to by the free pointer (1001)
02-1006), the necessary list is obtained, and then the list is inserted into the list pointed to by the in-use pointer (1002) described later.

In each list, an area that points to the next list and the data (BL
(K #) is held, and an area indicating where it is stored in the cache memory (142) is stored (1043-1045). Since the list pointed to by the free pointer is a chain of unused lists, the cache memory (142)
The value of the pointer to has no meaning.

Reference numeral 1007 is a hash list. The hash list (1007) is provided for the purpose of quickly searching whether the requested data is stored in the cache memory (142). By hashing using the block number BLK # of the requested disk drive as a key, it becomes possible to easily narrow down the candidates to be searched.

FIG. 13 shows the command queue list (50
The configuration of 6) is shown.

Reference numeral 1101 stores a pointer to a command queue corresponding to each local disk array controller (104, 105, 106, 107). 110
2 and 1103 are local disk array control devices (1
04) corresponding to new free tags (1110-1114)
And points to the list of used new tags (1115-1118). 1104 and 1105 are free new tags (1 corresponding to the local disk array controller (105).
119-1122) and new tags used (1123-112)
It points to the list in 7). 1106 and 1107
Points to a list of free new tags (1128-1132) and used new tags (1133-1136) corresponding to the local disk array controller (106).
1108 and 1109 are free new tags (1137-11) corresponding to the local disk array controller (107).
41) and the list of used new tags (1142-1145).

A pointer to the next list, a new tag value, and an input / output command (including a tag from the host) transferred from the host processor (101) are stored in each list. Free new tag (1110-111
4, 1119-1122, 1128-1132, 113
The value of the command area (cmd) in 7-1141) has no meaning. One of the input / output requests from the host processor (101) is stored in the used new tag list. New tag used (1115-1118, 112)
3-1127, 1133-1136, 1142-114
In step 5), the used new tag list is moved to the free new tag list at the end of command execution. The command queue list corresponding to each local disk array controller (104, 105, 106, 107) has a tag value that does not overlap. This command queue list (50
According to 6), it becomes possible to convert the tag transferred from the above-mentioned host processor (101) into a new tag.

FIG. 14 shows a flow chart of the command reception processing program (405). The command reception processing program (405) includes a command control unit (14
5) A microprogram stored in the memory (404) inside.

In step 1201, it is checked whether the cache can be secured. Step 1 if possible
If not, the process proceeds to step 1212, and command abnormal end processing is performed. Normally, there is no case where an area cannot be secured in the cache memory (142). The command abnormal termination process is executed when the cache memory (145) is inaccessible, for example, when a failure occurs.

At step 1202, a required area of the cache memory (145) is secured. Step 1203
Then, from the list pointed to by the free pointer (1001) of the cache management list (505), the required list is put in the list pointed by the in-use pointer (1002). In step 1204, I
It is checked whether or not there is a free list in the / O management list (503). If there is a free list, the process proceeds to step 1205, and if not, the process proceeds to step 1212 and abnormal termination processing is performed. This state means that there is no queue in the disk array control device (102) for receiving a command from the host processor (101).

At step 1205, a new list is added to the I / O management list (503). As a result, the command becomes the processing target of the disk array control device (102). In step 1206, the host ID is set to I /
It is stored in the area (903) of the O management list (503). In step 1208, the command reception time is set to I / O.
It is stored in the area (904) of the management list (503).
This time can be used for maintenance such as when the disk array controller (102) fails.

At step 1209, the command is input / output.
It is stored in the area (906) of the management list (503).
At this time, the area 907 of the I / O management list (503)
Also, the tag transferred from the host processor (101) is stored. This is referred to when the input / output request is completed, and is transferred to the host processor (101) together with the completion information and data if necessary to the host processor (101). In Step 1210, Step 1202
The command is transferred to the area of the cache memory (142) acquired in step. Here, the command is READ / WRI.
In the same manner, the instructions for the disk device such as the TE and the data for the WRITE request are also stored in the cache memory (14
It is stored in 2).

At step 1211, a flag indicating the completion of command reception is set in the status area (905) of the I / O management list (503). The execution result is stored in this status area (905) at the end of the command, but until then, the disk array controller (10
The execution status of the command in 2) is stored for each event and can be used as logging information when some kind of failure occurs.

FIG. 15 shows the end processing program (406).
The flowchart of FIG. The termination processing program (406) is a microprogram stored in the memory (404) in the command control unit (145).

In step 1301, it is determined whether the requested command is a READ request or a WRITE request.
This determination can be achieved by referring to the command area of the I / O management list (503). Judgment result, R
If it is an EAD request, the process proceeds to step 1306, and WRIT
If the request is E, the process proceeds to step 1302.

At step 1302, the contents of the status area (905) and the tag area (907) of the I / O management list (503) are merged. In step 1303, the data merged in step 1302 is transferred to the host processor (101) indicated by the host area (903) of the I / O management list (503). Then, it progresses to step 1304. Since the WRITE processing is already performed after writing the data in the disk drive or the cache area, the information to the host processor (101) is only whether or not the processing is normally completed.

Step 1306 is executed at the end of the READ processing request. The requested data has already been stored in the cache memory (142). In step 1306, this data and status area (90
5) and the contents of the tag area (907) are merged. In step 1307, the data merged in step 1306 is transferred to the host processor (101) indicated by the host area (903) of the I / O management list (503). Then, it progresses to step 1304.

At step 1304, the input / output request is deleted from the I / O management list (503).

FIG. 16 shows the global RAID controller (1
43). The global RAID controller (143) includes an adapter (1401) for controlling command and data transfer, a processor (1402), and a DMAC.
(1403), memory (1404), path switch (1
405) and a parity generation unit (1406).

Microprograms (1407, 1408, 14) for command control are stored in the memory (1404).
09, 1410, 1411, 1412) are stored. Specifically, the I / O request acceptance processing program (14
07), data recovery processing program (1408), I /
O request end processing program (1409), data allocation control program (1410), global parity control program (1411), and tag control program (1
412). These details will be described later. Micro programs (1407, 1408, 1409, 141
0, 1411, 1412) is a processor (1402)
Run on.

Data transfer with an external control unit is performed by the direct memory access controller DMAC (140
3) executes independently of the processor (1402).
The path switch (1405) refers to the header of the data transferred from the DMAC (1403) and performs an operation of distributing output signals. This operation will be described later in detail. The parity generation unit (1406) includes a parity buffer (1415) and the parity buffer (1415).
And a parity generator (1408) that creates parity data for the data stored (transferred) in the.

The main control flow of the global RAID controller (143) is as follows. First, data is received from the cache memory (142), and the data allocation control program (1410) divides the data if necessary. Next, for the divided data,
A header for identifying to which local disk array control device (104, 105, 106, 107) each data is transferred is added. Then, the path switch 1405 transfers the divided data to each local disk array control device (104, 105, 106, 107) according to the header.

FIG. 17 shows a flow chart of the I / O request acceptance processing program (1407).

In step 1501, a flag indicating that the global RAID control unit (143) has received a processing request is set in the status area (905) of the I / O management list (503). In step 1502, the command area (9) of the I / O management list (503)
The contents of 06) are transferred to the internal buffer. Then, the process proceeds to step 1503, and the data allocation control program (14
Control is transferred to 10).

FIG. 18 shows a data allocation control program (1
The flowchart of (410) is shown.

At step 1601, the request processing is REA.
Judge whether D or WRITE. If it is a READ processing request, the process proceeds to step 1602, and it is determined whether the requested data exists in the cache memory (142). If not, go to Step 1603 and set Global R
The AID control unit (143) determines whether the local disk array control device (104, 105, 106, 107) is in an output enable state.

The step 1603 will be described in detail. The global disk array controller (103)
Local disk array controller (104, 105, 1
06, 107), it is a control device located above
Due to the large striping size, the local disk array controller (104, 105, 106, 10
There are cases where it is better to buffer more data than 7). For example, four local disk array controllers (104, 105, 106, 107) are configured to store 4 KB of data in each of the four disk drives in units of 1 KB, and there are four global disk array controllers (103). In the case of full striping to the local disk array controller (104, 105, 106, 107) of 4 units, the global disk array controller (103) stores 4 KB * 4 units = 16 KB of data by buffering the four local units. Disk array control device (104, 105, 10
6, 107) can perform output processing simultaneously. In this case, in step 1603, it is determined whether the data can be output to all the local disk array controllers (104, 105, 106, 107) at the same time.

However, there are cases where such a determination is not necessary. In the above example, full striping is premised, but in some cases, like RAID 4 and 5, data striping is basically not performed. In that case, the data transferred from the host processor (101) is output to any one of the local disk array controllers (104, 105, 106, 107) without buffering. In another case, the virtual drive has a plurality of local disk array control devices (104, 1).
05, 106, 107). In this case, the local disk array controller (104, 10) is operated according to the normal operation of the local disk array controller (104, 105, 106, 107).
5, 106, 107), the global disk array controller (103)
Does not need to be buffered. Therefore, step 1
At 603, the virtual drive management table (502) and the local disk array management table (504) are referenced to determine how much data needs to be buffered.

At step 1605, the disk request address of the host command is converted into the address of the head local disk array controller (104, 105, 106, 107). This processing is performed by the host processor (101)
This is necessary because the drive recognized by does not match the location actually stored in the drive of the local disk array. The tables to be referred to are the virtual drive management table (502) and the local disk array management table (504).

For example, the requested disk address from the host processor (101) is 12K from the 32KB from the beginning of VOL2 of the virtual drive management table (502).
B data output. In that case, the method of obtaining the start address of the local disk array control device (104, 105, 106, 107) is as follows (1) to (4).

(1) By referring to the virtual drive management table (502) in FIG. 9, how the virtual drives are arranged in the local disk array controller (104, 105, 106, 107) is obtained. As a result, V
OL2 is a local disk array controller (104, 1
05, 106, 107).

(2) The partition 2 of the local disk array controller (104, 105, 106, 107) is the local disk array management table (5) of FIG.
04), columns 605, 606, and 608 indicate RAID level 3 and parity number 1, respectively, and partition 2 is for each local disk array controller (10).
4, 105, 106, 107).

(3) From the columns 604 and 607 of the local disk array management table (504) of FIG. 8, the local disk array control units (104, 105, 10) are selected.
6, 107), the number of drives is 5, and the striping size is 4 KB. Number of parity is 1
Therefore, each local disk array controller (104, 105, 106, 107) has 4 KB *
(5 units-1 unit) = 16 KB will be stored.

(4) As a result, the requested address is 32.
The physical address of the KBth is 32 KB / 16 KB + 1
= The third local disk array controller (106)
The 1st GB becomes the start address.

Next, in step 1605, it is determined whether the request command is READ or WRITE. As a result, R
If it is EAD, the process proceeds to step 1606, and if it is WRITE, the process proceeds to step 1611.

At step 1606, a command is generated to issue an input / output request to the local disk array controller (104, 105, 106, 107). At this time, each local disk array controller (1
04, 105, 106, 107) specifies the striping size of the column 709 of FIG.
In step 1607, the tag control program (1412) is executed to generate a new tag. Details of the tag control program (1412) will be described later.

In step 1608, step 1606
And the command generated in step 1607 and the new tag are merged, and a header for identifying which local disk array controller (104, 105, 106, 107) the request is made to is added, and the path switch ( 140
Data is transferred to 5). At step 1609, the next local disk array controller is selected. This operation can be performed in the same manner as the calculation performed when obtaining the physical address. This is possible by simply recalculating as an address obtained by adding the striping size in column 709 of FIG. 9 to the storage request address.

Next, in step 1610, it is determined whether or not all the data has been processed. If not processed, the process is repeated from step 1606, and if the process is completed, this process is ended.

If it is determined to be WRITE in step 1605, the process proceeds to step 1611. Step 1611
Means execution of the global parity control program. This process will be described in detail later.

At step 1612, a command is generated to issue an input / output request to the local disk array controller (104, 105, 106, 107). At this time, each local disk array controller (1
04, 105, 106, 107) specifies the striping size of the column 709 of FIG.
In step 1613, the tag control program (1412) is executed to generate a new tag. Details of the tag control program (1412) will be described later.

Next, in step 1614, step 1
Similar to 608, the local disk array controller (1
04, 105, 106, 107). Step 1615 is the same process as step 1609, and step 1616 is the same process as step 1610.

FIG. 19 illustrates the operation of the path switch (1405).

In this example, four commands (1709,
1710, 1711, 1712) is the selector (141
4) shows how distributed transfer is performed. Each command includes a command field (1701, 1703, 1705, 1707) such as READ / WRITE and a path number (1702, 1704, 1706) indicating a transfer destination.
1708) and.

For example, the packet 1712 has a path number of 1, the packet 1711 has a path number of 2, and the packet 171
0 has a path number of 3, and packet 1709 has a path number of 4. The selector (1414) refers to this path number to determine to which path the data following the path number should be transferred, and performs path switching. For example, if the pass number is 1, the selector (141
4) selects path number 1 and transfers the data (171
3). Similarly, packets 1711, 1710, 1709
Also, by the selector (1414), 1714, 171
5, 1716 are distributed and transferred.

FIG. 20 is a flow chart of the global parity control program (1411).

In step 1801, the parity position is calculated from the virtual drive management table (502).
For example, the requested disk address is 1 from the 32 KB from the beginning of VOL2 of the virtual drive management table (502).
2KB data output. The method of obtaining the parity position in that case is as described in (1) to (4) below.

(1) Virtual drive management table (50
2), the local disk array controller (1
04, 105, 106, 107) and how virtual drives are arranged. As a result, VOL2
Is a local disk array controller (104, 105,
106, 107).

(2) The partition 2 of the local disk array control device (104, 105, 106, 107) has a local disk array management table (504).
From columns 605, 606 and 608 of
ID level 3, parity number 1 and partition 2
Indicates each local disk array controller (104, 10
It can be seen that it starts from the first GB (5, 106, 107).

(3) Since the global RAID level is 3, it can be seen that the parity is the allocated final local disk array (107).

(4) As a result, the physical address of the parity is from the first GB of the local disk array (107).

The above parity position calculation is an example, and the local disk array controller (104, 105, 10) is used.
6, 107) and the RAID level of the global disk array controller (103), but can be easily realized by applying the arrangement rule of each RAID level to the above-mentioned calculation.

Next, in step 1802, it is determined whether or not the data currently stored in the drive is required for the parity calculation. That is, the entire parity may not be updated depending on the RAID level. For example R
AID level 5 and the like. The parity is not limited to RAID level 5, and one or more than one parity is generated from a plurality of data. At this time, if all the data related to a certain parity are updated at the same time, a completely new parity can be generated from all the new data. However, in the case of partial data update such as RAID level 5, first, the information of the past data at the location to be updated must be deleted from the parity, and then the parity of the new data must be created.

If it is determined from the determination result of step 1802 that the old data is not needed, the process proceeds to step 1806. If not, the process proceeds to step 1803. In step 1803, an input command to the local disk array controller (104, 105, 106, 107) is generated to read the old data. This process is equivalent to step 1606 of FIG. As the input address, the address obtained in step 1801 is used. In step 1804, the tag control program (1412) is executed to generate a new tag. This process will be described in detail later. In step 1805, the local disk array controller (104, 105, 106, 10
Issue an input request to 7). This processing is equivalent to step 1608 and the like in FIG.

At step 1806, the data is transferred from the cache area to the parity generator (1406) by designating the striping size of the global disk array controller (103). Step 1 if necessary
The old data acquired in 805 is also transferred at the same time. The striping size of the global disk array controller (103) can be obtained by referring to the striping size in the column 709 of FIG. This operation will be described later in detail.

Next, in step 1807, the generated parity is read from the parity generator (1406). In step 1808, a command for writing the new parity acquired in step 1807 is generated for the parity position calculated in step 1801.
In step 1809, the tag control program (1412) is executed to acquire the tag for the command generated in step 1808. In step 1810, an output process is performed to write a new parity in the parity position obtained in step 1801.

FIG. 21 shows how global parity is generated.

Reference numeral 1901 is a data group for generating a parity, and 1902 is a striping size of the global disk array controller (103). By transferring these two pieces of data to the parity generation unit (1406), the data group (1901) is first divided into striping sizes of the global disk array control device (103) (1903, 1904, 1905), and each is exclusive. Logical OR is taken (1906, 1907). Thereby, global parity is generated.

In FIG. 22, when the input / output command transferred from the host processor (101) is the disk device (13
No. 3,134,135,136), it shows how it changes before being stored or read out.

From the host processor (101), data (2001), data length (2002), command (2)
003), host identifier (2004), tag (200
5) and the virtual drive identifier (2006) are transferred as one packet. The global disk array control device (103) which received it receives the local disk array control devices (104, 105, 106, 10).
A plurality of commands are generated in order to issue the input / output request for 7).

As shown in 2007-2012, each of them is a packet of data, data length, command, host identifier, tag, and local disk array identifier (path number). Here, the packet format is the same as that of 2001-2006, but the content is different.

For example, the data (2007) is smaller than the data (2001) because it is divided. Accordingly, the data length (2008) is also set to a value shorter than the data length (2002). Command (2009)
Changes the storage or read position to the address of the local disk array. The host identifier (2010) is
Unlike the identifier of the host processor (101), this is the identifier of the global disk array controller (103) in this case. The tag (2011) is not a tag transferred from the host processor (101) but a new tag. The reason for this has already been mentioned. A local disk array identifier (path number) that is invisible to the host processor (101) is stored in 2012.

Local Disk Array Controller (10
4, 105, 106, 107), 2013-20
As shown in 17, the data, the data length, the command, the host identifier, and the drive number are transferred as one packet to the drive. The division is performed in the same manner as the change of the command in the global disk array controller (103) described above.

FIG. 23 shows the tag control program (141
The flowchart of 2) is shown. As described above, this tag is the tag transferred from the host processor (101) from the global disk array controller (103) to the local disk array controller (104, 105,
This is because it cannot be used for input / output with respect to (106, 107), and therefore tag conversion is necessary.

At step 2101, the local disk array controller (104, 105, 10) for input / output.
Command queue list (50
Free tag pointers (1102, 1104, 1) in 6)
106, 1108). The free tag pointer indicates the local disk array controller (104, 10).
5, 106, 107) and each local disk array control device (104, 105, 106, 1)
07) tags not used for input / output are chained in a list format (FIG. 13). Step 210
In step 2, one tag is acquired from this list.

Next, in step 2103, the acquired tag is reconnected to the in-use list. As a result, the acquired tag is not used by another request, and the same local disk array control device (104, 105, 1)
No duplicate tags in 06,107) are used. In step 2104, the acquired tag is stored in the new tag area (908) in the I / O management list (503). This is done to associate with the tag transferred from the host processor (101). As a result, when the input / output is completed, when the tag transferred from the host processor (101) is returned to the host processor (101), it becomes possible to easily retrieve it. Step 21
In 05, the tag is added to the command and stored in the command storage area in the command queue list (506).

FIG. 24 shows the data recovery control program (1
408) is shown.

The disk array device is provided with data recovery means. In the present invention, parity is hierarchized in order to make the disk array more reliable than the normal disk array. Parity managed by the local disk array controller (104, 105, 106, 107) and the global disk array controller (1
03) manages global parity. The global parity is the parity across a plurality of local disk array controllers (104, 105, 106, 107).

Local Disk Array Controller (10
4, 105, 106, 107) manage the parity
The local disk array controller (104, 10
5,106,107) only.
Therefore, the local disk array controller (104, 1
05, 106, 107), if any of the disk drives in
The data on the failed drive can be recovered.

The global parity is defined by a plurality of local disk array controllers (104, 105, 106, 10).
Since it is created over 7), it can be used even when the above-mentioned single disk drive fails, but more than that, the local disk array controller (104, 1)
(05, 106, 107) itself can be used when it fails.

The flowchart of FIG. 24 shows an operation in which failure recovery is layered.

At step 2201, it is judged whether the recovery processing of the local disk array control device (104, 105, 106, 107) is possible. Various factors can be considered for this determination. For example, in the event of failure of the failure recovery mechanism or failure beyond the range that can be recovered by parity, the local disk array controller (104, 105, 10)
6,107) Failure recovery cannot be done by itself. In this case, the process proceeds to step 2202.

At step 2202, it is judged whether or not there is a failure in the parity area. If it is a parity failure, it is not a failure of the data portion, so the recovery processing is not performed here. In step 2203, data other than the failed local disk array is acquired. In this case, the parity data is also included. In Step 2204, Step 2
From the data or parity acquired in 203, the data in the failed portion is recovered.

Although not described here, global parity may also exist. this is,
The global disk array controller (103) is the local disk array controller (104, 105, 106,
107) when the global parity data is stored in the local disk array controller (104, 10).
5, 106, 107), the parity is created / stored in the global parity data as in the normal data. In the present invention, since the parity information is also layered in this way, high reliability can be obtained.

FIG. 25 shows a flow chart of the I / O request end processing program (1409).

In step 2301, the command queue list (506) is searched for a list corresponding to the request to be terminated. In step 2302, the I / O management list (503) is searched in the same manner as in step 2301.
In step 2304, the end state transferred from the local disk array control device (104, 105, 106, 107) is stored in the status area (905) of the I / O management list (503).

In step 2304, the list searched in step 2301 is deleted from the command queue list (506), and the unused list is replaced. This allows
The tag used in this processing can be used for other input / output. In step 2305, the host processor (1
01) from the I / O management list (50
The process is ended by taking out from the tag area (907) of 3) and notifying the command control unit (145).

Next, the details of the local disk array control device (104, 105, 106, 107) will be described.
Among them, the command control unit (113, 114, 115,
116) is the global disk array controller 103
Command control unit 141. Also, the interface control unit (109, 110, 111, 112)
This is equivalent to the interface control unit 108 of the global disk array control device 103. Also, the interface controllers (129, 130, 131, 132) are
This is equivalent to the interface controller 144 of the global disk array controller 103.

FIG. 26 shows the configuration of the RAID controller (125, 126, 127, 128) of the local disk array controller (104, 105, 106, 107). Note that, in FIG. 26, portions common to the respective portions of the global RAID control unit 143 shown in FIG. 16 are indicated by the same numbers.

RAID controller (125, 126, 12
7, 128) are an adapter (1401), a processor (1402), a DMA that controls the transfer of commands and data.
The C (1403), the memory (1404), the path switch (1405), and the parity generation unit (1406).

Microprograms (1407, 2404, 24) for command control are stored in the memory (1404).
01, 2402, 2403) are stored. Specifically, the I / O request acceptance processing program (1407), the data recovery processing program (2404), the I / O request end processing program (2401), the data allocation control program (2403), and the parity control program (24
02).

I / O request acceptance processing program (140
Since 7) is the same as that in FIG. 17, its explanation is omitted. I /
O request end processing program (2401), data allocation control program (2403), parity control program (2402), and data recovery processing program (24
Details of 04) will be described later. Micro Program (14
07, 2404, 2140, 2402, 2403) are executed by the processor (1402).

Data transfer with an external control unit is performed by the DMAC.
(1403) executes independently of the processor (1402). The path switch (1405) is connected to the DMAC (140
The header of the data transferred from 3) is referred to, and the operation of distributing the output signal is performed. Parity generator (1606)
Is composed of a parity buffer (1415) and a parity generator (1408) that creates parity data of the data stored (transferred) in the parity buffer (1415).

RAID controller (125, 126, 12
The large control flow of (7, 128) is as follows. First, the global disk array controller (1
03), and the data allocation control program (2403) divides the data if necessary. Next, for each of the divided data, which one of the disk drives (137, 138, 139, 140)
Add a header to identify whether to transfer to. After that, the path switch (1405) transfers the divided data to each disk drive (137, 138, 139, 140) according to the header.

FIG. 27 shows the control memories (117, 118,
119, 120). Reference numeral 2502 is an I / O management list, and 2503 is a cache management list.

The I / O management list (2502) is a table for managing the input / output requests received from the global disk array controller (103) together with the input / output states and the like until the input / output becomes the environment. Cache management list (25
03 is a cache memory (121, 122, 12)
3, 124) and the address thereof is stored. The I / O management list (2502) and the cache management list (2503) are the I / O management list 503 (FIG. 11) and the cache management list 505 of FIG.
It is a list equivalent to (FIG. 12).

FIG. 28 shows a flow chart of the I / O request end processing program (2401).

At step 2601, the list of input / output requests to be ended is searched from the I / O management list (2502). In step 2602, the I / O management list (25
02) in the status area of the disk device (133, 1
34, 135, 136), the processing is terminated by storing the termination state transferred from the above.

FIG. 29 shows a data allocation control program (2
The flowchart of (403) is shown.

At step 2701, the request processing is REA.
Judge whether D or WRITE. If it is a READ processing request, the process proceeds to step 2702, and it is determined whether the requested data exists in the cache memory (142).
When the WRITE processing request is made in step 2701, the process proceeds to step 2703, and the RAID controller (125, 12
6, 127, 128) are disk devices (133, 13
4, 135, 136), it is determined whether output is possible.

At step 2704, the disk request address of the global disk array controller (103) is set to the head disk device (133, 134, 135, 136).
Address. In step 2705, it is determined whether the request command is READ or WRITE. As a result, if it is READ, the process proceeds to step 2706 and the WRI
If it is TE, the process proceeds to step 2709.

At step 2706, an input / output request is issued to the disk drive (137, 138, 139, 140). In step 2707, the next disk drive (137, 138, 139, 140) is selected. In step 2708, it is determined whether or not all data processing has been completed, and if not completed, step 2706.
Repeat from.

Step 2709 is equivalent to step 2706. Step 2710 is equivalent to step 2707. Step 2711 is equivalent to step 2708. In step 2712, the parity control program (2402) is executed.

FIG. 30 shows the parity control program (24
02) is a flow chart.

At step 2801, the RAID level,
The parity position is calculated from the number of drives and the requested disk address. In step 2802, it is determined whether old data is needed for updating the parity. From the determination result of step 2802, if old data is not needed, the process proceeds to step 2804, and if not, step 2
Proceed to 803.

At step 2803, the disk drive (137, 138, 13) is read in order to read the old data.
9, 140) to issue the old data input request.
As the input address, the address obtained in step 2801 is used.

At step 2804, data is transferred from the cache area to the parity generation unit (1406).
If necessary, the old data acquired in step 2803 is also transferred at the same time. In step 2805, the generated parity is read from the parity generation unit (1406). In step 2806, output processing is performed to write a new parity in the parity position obtained in step 2801.

FIG. 31 shows the data recovery control program (2
404) is shown. The disk array device is provided with data recovery means. Therefore, the local disk array controller (104, 10
If any disk drive in 5, 106, 107) fails, its parity can be used to recover the data in the failed drive.

First, in step 2901, it is determined whether or not there is a failure in the parity area. If it is a parity failure, it is not a failure of the data portion, so the recovery processing is not performed here. In step 2902, data other than the disk drive in which the failure has occurred is acquired. In this case, the parity data is also included. In Step 2903, Step 29
The data in the faulty part is recovered from the data or parity acquired in 02.

FIG. 32 shows a flow chart of the command reception processing program. The command reception processing program is a command control unit (113, 114, 115, 1
16) A microprogram stored in the memory.

In step 3001, it is checked whether the cache can be secured. If possible, the process proceeds to step 3002, and if not, the process proceeds to step 3013 to execute command abnormal end processing. In step 3002, the cache memory (121, 122, 123, 12
Reserve the required area in 4). In step 3003,
From the list pointed to by the free pointer (1001) of the cache management list (2503), put it in the list pointed to by the in-use pointer (1002).

In step 3004, it is checked whether or not there is a free list in the I / O management list (2502). If there is a free list, the process proceeds to step 3005, and if not, the process proceeds to step 3013 and abnormal termination processing is performed.

At step 3005, a new I / O management list (2502) is added. In step 3006, the host ID is stored in the area (903) of the I / O management list (2502). In step 3008, the command reception time is stored in the area (904) of the I / O management list (2502). This time can be used for maintenance such as when the disk array controller (102) fails.

At step 3009, the command is input / output.
It is stored in the area (906) of the management list (2502). In step 3010, the cache memory (121, 122, 123, 12 acquired in step 3002
4) Transfer the command to the area. Here, the command is R
If an instruction to the disk device such as EAD / WRITE and a WRITE request, the data is also stored in the cache memory (142).

At step 3011, a flag indicating the completion of command reception is set in the status area (905) of the I / O management list (2502). This status area (905) stores the execution result at the end of the command, but until then, the execution status of the command in the local disk array control device (104, 105, 106, 107) is stored for each event. , It can be used as logging information when some kind of failure occurs.

FIG. 33 shows a flowchart of the end processing program. The termination processing program is a microprogram stored in the memory in the command control unit (113, 114, 115, 116).

In step 1301, it is determined whether the requested command is a READ request or a WRITE request.
This determination can be achieved by referring to the command area of the I / O management list (2502). Judgment result,
If it is a READ request, the process proceeds to step 3103, and WRI
If it is a TE request, the process proceeds to step 3102.

At step 3102, the status area (905) of the I / O management list (2502) is transferred to the global disk array controller (103). Then, it progresses to step 3105.

Step 3103 is executed at the end of the READ processing request. Cache memory (121, 12
2, 123, 124), the requested data has already been stored. In step 3103, this data and the contents of the status area (905) are merged. In step 3104, the data merged in step 3103 is transferred to the global disk array controller (103) indicated by the host area (903) in the I / O management list (2502). Then, it progresses to step 3105.

At step 3105, the input / output request is deleted from the I / O management list (2502).

FIG. 34 shows an example of the failure recovery operation in this embodiment. An example of the failure recovery operation according to the present invention will be described with reference to this figure.

In this example, one of the disk drives managed by the local disk array controller (104) fails, and the local disk array controller (1)
05) when a failure occurs, the host processor (10
It shows a recovery operation when a data read request is issued from 1).

First, the local disk array controller (104) reads data from a drive other than the drive in which the failure has occurred, and restores the data in the drive in which the failure has occurred (3204). The restored data and the data of the normal drive are merged (3203) and transferred to the global disk array controller (103).

Local Disk Array Controller (10
5) transfers the requested data (3205) to the global disk array controller (103) as it is.
The global disk array control device (103) judges that the local disk array control device (105) cannot repair independently, and performs data repair using global parity.

Therefore, the global parity (3206) required for recovery is read from the local disk array controller (107) in which the global parity is stored. The global parity (3206) and the data of the local disk array from which the data was read normally,
In the global disk array controller (103), the local disk array controller (10) in which the failure has occurred
The data of 5) is restored (3202). The restored data is merged with the data of the local disk array from which the data was read normally (3201) and transferred to the host processor (101).

As described above, according to the present invention, the processing of the host processor (101) is not interrupted not only when the drive fails but also when the disk array itself fails.

FIG. 35 shows an example of storing parity data in the disk array device according to the present invention. In the figure,
P is a local disk array controller (104, 10
5, 106, 107), and GP indicates the global disk array controller (10
3 shows the parity generated by 3).

Reference numerals 3301 to 3304 indicate the parity when the RAID level of the local disk array control device (104, 105, 106, 107) is 3 or 4.
On the other hand, 3300 is the parity generated by the global disk array controller (103), and at this time, the RAID of the global disk array controller (103)
The level is 3 or 4.

Similarly, 3309-3312 are local disk array controllers (104, 105, 106, 1).
07) shows the parity when the RAID level is 3 or 4. Reference numeral 3305-3308 is a parity generated by the global disk array control device (103), and at this time, R of the global disk array control device (103).
The AID level is 5. In this way, not only the data but also the parity is stored across the plurality of local disk array controllers (104, 105, 106, 107).

The same applies to the global parity of 3313. At this time, the local disk array controller (10
4, 105, 106, 107) have a RAID level of 5.

Similar to FIG. 35, FIG. 36 shows an example of storing parity data. FIG. 35 shows an example in which the parity is generated / stored in both the global disk array control device (103) and the local disk array control device (104, 105, 106, 107), so that extremely high reliability can be obtained. It was FIG. 36 is an example showing a usage method in which a large capacity is prioritized over reliability.

The numeral 3401 is the only parity in this disk array device. There is no parity in the other local disk array controllers (104, 105, 106). By doing so, it is possible to configure a disk array that prioritizes capacity.

As another example similar to this, global parity can be stored in all the disk drives of the local disk array controller (107). Three
Reference numeral 402 is the RAID 5 global parity. However, other local disk array controllers (104,
No parity exists in (105, 106).

Even with global parity of the same RAID 5, it is possible to disperse and arrange the local disk array control devices (104, 105, 106, 107) like 3403-3406.

[Embodiment 2] Next, a second embodiment of the present invention will be described.

FIG. 37 shows an overall configuration diagram of the second embodiment of the present invention. The same parts as those in FIG. 1 are indicated by the same numbers. A big difference from FIG. 1 is that the global disk array controller (3513) and the local disk array controller (3502, 3503, 3504, 34).
05), the common bus (3506) is being used. Control memory (3511), global RAI
D control unit (3501) and command control unit (350
7-3510) will be described in detail later, but the other parts are the same as those in FIG.

FIG. 38 shows the global RAID controller (3
501). The basic configuration is shown in FIG.
However, the microprogram (3601) and the multicast control unit (3602) are different.

Global RAID Controller (3501)
Receives an input / output request from the host processor (101) and transfers data to the local disk array controller (35
02,3503,3504,3405). At this time, the global RAID control unit (3501)
Are all local disk array control devices (3502, 3503, 3504, 340) connected to the common bus (3506) by the multicast control unit (3602).
The same command is transferred with 5) as the transfer destination.

FIG. 39 shows the structure of the control memory (3511) in the local disk array controller (3502, 3503, 3504, 3405).

Tables and lists (502, 503, 50
4, 505) is equivalent to that of FIG. 7, but a local disk array identifier (3702) is added. The local disk array identifier (3702) is the local disk array controller (3502, 3503, 3504).
A different identifier is set for each 3405). The local disk array identifier (3702) is the local disk array control device (3502, 3503, 3504, 3).
405) to determine whether to execute the command or data transferred from the global RAID controller (3501).

FIG. 40 shows a flowchart of the command reception processing program. The command reception processing program includes command control units (3507, 3508, 350).
9, 3510) is a microprogram stored in the memory.

At step 3801, the control memory (35
11) Local disk array identifier (3702)
And the command transferred, it is determined whether or not the content is to be executed by the local disk array. This decision is
This is performed using the local disk array identifier (3702) and the virtual drive management table (502). For example, when there is an input / output request for the block m of the virtual drive n from the global RAID control unit (3501), the following determination is made.

(1) Virtual drive management table (50
From 2), the identifier of the local disk array in which the virtual drive n and the block m are stored is obtained. (2) Compare the obtained identifier with the local disk array identifier (3702) in the control memory (3511),
Determine whether the local disk array should be executed.

With the above operation, the determination in step 3801 becomes possible. If it is a command to be executed in its own local disk array, the process proceeds to step 3802, and if it is the total output, the process ends.

At step 3802, it is checked whether the cache can be secured. If possible, the process moves to step 3803, and if not, the process moves to step 3814 to execute command abnormal end processing. In step 3803, a required area of the cache memory (121-124) is secured. In step 3804, the required list is put into the list pointed to by the in-use pointer (1002) from the list pointed to by the free pointer (1001) of the cache management list (505).

Next, in step 3805, it is checked whether or not there is a free list in the I / O management list (503). If there is a free list, step 3806.
If not, the process proceeds to step 3814 to perform abnormal termination processing.

At step 3806, a new list is added to the I / O management list (503). Step 380
In 7, the host ID is stored in the area (903) of the I / O management list (503). In step 3809, the command reception time is stored in the area (904) of the I / O management list (503). This time can be used for maintenance such as when the disk array controller (102) fails. In step 3810, the command is sent to the I / O management list (503) area (906).
To store.

In Step 3811, Step 3803
The command is transferred to the cache memory (121-124) area acquired in step. Here, the command is READ / W
Instructions for disk devices such as RITE and WRIT
If it is an E request, the data is also cache memory (12
1-124). In step 3812, I
/ O management list (503) status area (90
A flag indicating the completion of command reception is set in 5). In the status area (905), the execution result is stored when the command ends, but until then, the execution state of the command in the disk array control device (102) is stored for each event, and logging is performed when some kind of failure occurs. It can be used as information.

FIG. 41 shows the data allocation control program (1
410) shows an operation flowchart.

At step 3901, the request processing is REA.
Judge whether D or WRITE. If it is a READ processing request, the process proceeds to step 1602, and it is determined whether the requested data exists in the cache memory (142).
If it is a WRITE processing request in step 3901, the process proceeds to step 3903. At step 3903, the request from the host processor (101) is directly transmitted to the local disk array controller (3502, 3503, 3504,).
3405).

FIG. 42 shows an example of the operation of the second embodiment.

Global Disk Array Controller (35
13) to data 4002, input / output data length 400
3, the input / output start address 4004, the virtual drive identifier 4005, and the command 4006 are included in one packet for all local disk array control devices (35
02,3503,3504,3405) (4001).

Local Disk Array Controller (350
2, 3503, 3504, 3405), as a result of the execution determination, 3502 and 3505 have an input / output that is unrelated to the local disk array control device itself, and the command is canceled. Local disk array controller (350
3, 3504) interprets the command transferred from the global disk array controller (3513) and executes it (4007, 4008).

By such processing, the local disk array control device (3502, 3503, 3504, 34)
It is possible to improve the connectivity of 05). That is,
In the global disk array controller (3513), it is no longer necessary to be aware of the local disk array controller (3502, 3503, 3504, 3405) in input / output control. Therefore, even if a new local disk array controller is added, it is not necessary to change the global disk array controller (3513). Further, the local disk array control device can be added by simply changing the local disk array identifier in the control memory (3507-3510).

[0218]

According to the present invention, even if a large number of disk devices are connected, high-speed input / output is possible, the cost is not increased, the disk management is easy, and the reliability is lowered. Absent.

[Brief description of drawings]

FIG. 1 is a general view of an embodiment of a disk array according to the present invention.

FIG. 2 shows an operation outline of the first embodiment according to the present invention.

FIG. 3 shows an operation outline of the first embodiment according to the present invention.

FIG. 4 shows an operation outline of the first embodiment according to the present invention.

FIG. 5 shows an operation outline of the first embodiment according to the present invention.

FIG. 6 shows a configuration of a command control unit.

FIG. 7 shows a configuration of a control memory.

FIG. 8 shows a configuration of a local disk array management table.

FIG. 9 shows a configuration of a virtual drive management table.

FIG. 10 shows an outline of a virtual drive.

FIG. 11 shows a structure of an I / O management list.

FIG. 12 shows the structure of a cache management list.

FIG. 13 shows the structure of a command queue list.

FIG. 14 shows a flow of a command reception processing program.

FIG. 15 shows a flow of an end processing program.

FIG. 16 shows a configuration of a global RAID control unit.

FIG. 17 shows a flow of an I / O request acceptance processing program.

FIG. 18 shows a flow of a data allocation control program.

FIG. 19 shows the operation of the pass switch.

FIG. 20 shows a flow of a global parity control program.

FIG. 21 shows an outline of generation of global parity.

FIG. 22 shows the flow of a command packet.

FIG. 23 shows a flow of a tag control program.

FIG. 24 shows a flow of a data recovery control program.

FIG. 25 shows a flow of an I / O request end program.

FIG. 26 shows a configuration of a RAID controller of the local disk array.

FIG. 27 shows a configuration of a control memory of the local disk array.

FIG. 28 shows a flow of an I / O request end program for the local disk array.

FIG. 29 shows a flow of a data arrangement control program for a local disk array.

FIG. 30 shows a flow of a parity control program for a local disk array.

FIG. 31 shows a flow of a data recovery control program for a local disk array.

FIG. 32 shows a flow of a command reception processing program of the local disk array.

FIG. 33 shows a flow of a termination processing program for a local disk array.

FIG. 34 shows the operation of a data recovery process according to the present invention.

FIG. 35 shows an example of data storage according to the present invention.

FIG. 36 shows an example of data storage according to the present invention.

FIG. 37 shows an overall configuration diagram of a second embodiment according to the present invention.

FIG. 38 shows a configuration of a global RAID control unit according to the second embodiment.

FIG. 39 shows the configuration of the control memory of the second embodiment.

FIG. 40 shows a flow of the command reception processing program of the second embodiment.

FIG. 41 shows a flow of the data allocation control program of the second embodiment.

FIG. 42 shows an operation example of the second embodiment.

[Explanation of symbols]

101 ... Host processor, 102 ... Disk array control device, 108 ... Interface control unit, 133, 13
4, 135, 136 ... Disk device, 137, 138,
139, 140 ... Disk drives, 103 ... Global disk array controller, 104, 105, 106,
107 ... Local disk array control device, 141 ... Command control unit, 145 ... Control memory, 142 ... Cache memory, 143 ... Global RAID control unit, 144
... interface controller, 402 ... processor, 502 ... virtual drive management table, 503 ... I
I / O management list, 504 ... Local disk array management table, 505 ... Cache management list, 506 ... Command queue list, 405 ... Command reception processing program, 406 ... End processing program, 1407 ... I / O
Request acceptance processing program, 1408 ... Data recovery processing program, 1409 ... I / O request end processing program,
1410 ... Data allocation control program, 1411 ... Global parity control program, 1412 ... Tag control program, 1405 ... Path switch.

Claims (13)

[Claims] 1. A means for dividing the data transferred from the host processor into a predetermined number of groups and outputting the data of each group in parallel, and the data of each group output in parallel is input and stored in group units. A disk array device comprising a plurality of local disk array devices. 2. A plurality of local disk array devices and which of the plurality of local disk array devices to output the data transferred from the host processor is selected, and the selected disk array device is provided with the local disk array device. A disk array device comprising means for outputting data. 3. A means for dividing data transferred from a host processor into a predetermined number of groups and outputting the data of each group in parallel; and a parity between the plurality of groups to generate data for each group. It comprises means for outputting in parallel, a plurality of local disk array devices for respectively inputting and storing the data of each group output in parallel in group units, and a local disk array device for inputting and storing the parity. A disk array device characterized by the above. 4. A disk array device comprising a plurality of disk drives, wherein the data transferred from a host processor and the parity generated from the data are stored in the disk drives, wherein the plurality of disk drives are respectively It is divided into a plurality of logical groups including a plurality of disk drives, and one or more of the logical groups are selected based on the address and data length in the volume transferred from the host processor, and the selected logical group is selected. A means for transferring the data to be stored, and a parity provided in each logical group for generating parity in the logical group from the transferred data, and generating the data in a plurality of disk drives in the own logical group. And a means for storing parity and Array device. 5. A disk array device including a plurality of disk drives, wherein the data transferred from a host processor and the parity generated from the data are stored in the disk drive, wherein each of the plurality of disk drives is a disk array device. It is divided into a plurality of logical groups including a plurality of disk drives, and one or more of the logical groups are selected based on the address and data length in the volume transferred from the host processor, and the selected logical group is selected. A means for transferring data to be stored, and parity between logical groups is generated using the data transferred to each of the selected logical groups, a logical group different from the selected logical group is selected, and the selected logical group is selected. As the data to be stored in the group, the logical group And a parity provided in each logical group for generating parity in the logical group from the transferred data to be stored, and the data and the generated parity in a plurality of disk drives in the own logical group. And a means for storing the disk array device. 6. The disk array device according to claim 4 or 5, wherein, with respect to other disk drives other than the disk drive in which parity is stored in the logical group, the storage unit for each disk drive is different. A disk array device characterized in that the addresses in the volume are successively allocated, and further, the addresses in the volume are horizontally allocated sequentially to the plurality of logical groups. 7. The disk array device according to claim 6, wherein a continuous portion of the addresses in the volume in the logical group is used as a minimum storage unit of the logical group, and data transferred from a host processor is stored in the minimum storage. A disk array device characterized in that it is horizontally divided into a plurality of logical groups for each unit and is transferred to each logical group. 8. The disk array device according to claim 7, wherein, in each of the logical groups, the data transferred in the minimum storage unit is horizontally transferred to a plurality of disk drives for each storage unit of the disk drive. A disk array device characterized by storing. 9. The disk array device according to claim 8, wherein data satisfying the minimum storage unit of the logical group is sequentially transferred to the logical group corresponding to the data. Disk array device. 10. The disk array device according to claim 6, wherein when data of a continuous portion of the address in the volume in the logical group is transferred from the host processor, parity is generated from the transferred data. Then, the data and the parity are stored in parallel in the disk drives in the logical group. 11. The disk array device according to claim 4, wherein when any disk drive in the logical group fails, the parity and the failed disk drive stored in the logical group are stored. A disk array device, wherein data of the failed disk drive is generated from data of a disk drive other than the above. 12. The disk array device according to claim 5, wherein when a failure occurs in any of the logical groups, the inter-logical group parity and data of a logical group other than the failed logical group are stored. From the disk array device, the data of the logical group in which the failure has occurred is generated. 13. A disk array device including a plurality of disk drives, wherein the data transferred from a host processor and the parity generated from the data are stored in the disk drives, wherein the plurality of disk drives are respectively It is divided into a plurality of logical groups including a plurality of disk drives, and a logical group control device is provided for each logical group. The logical group control device is configured such that the I / O commands transferred from the host processor are Determines whether it is an I / O request for the logical group corresponding to the local device, and if it is an I / O request for the logical group corresponding to the local device, executes the I / O command, otherwise cancels the I / O command. A disk array device comprising means for performing.