JP2854471B2 - Disk array device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a disk array device in which a plurality of disk devices are accessed in parallel to perform data input / output processing, and more particularly to a disk array device having a plurality of ranks and a dual port configuration. 2. Description of the Related Art As an external storage device of a computer system, disk devices such as a magnetic disk device and an optical disk device having features such as non-volatility of recording, large capacity, and high speed of data transfer are widely used. Demands for disk devices are high-speed data transfer, high reliability, large capacity, and low price. Disk array devices have been attracting attention as satisfying these requirements. The disk array device is a device for arranging a plurality of small disk devices, distributing data therein, recording data, and accessing in parallel.

When data is transferred to a plurality of disk devices in parallel by a disk array device, high-speed data transfer can be performed twice as many as the number of disks as compared with a single disk device. In addition, by recording redundant information such as parity data in addition to the data, it is possible to detect and correct data errors caused by failure of the disk device, and to duplicate the contents of the disk device. The same high reliability as that of the method of recording data can be realized at a lower price than duplication.

[0003]

2. Description of the Related Art Conventionally, David A. Patterson of the University of California, Berkeley.
et al. disclose a paper that evaluates a disk array device that accesses a large amount of data to many disks at high speed and realizes data redundancy at the time of disk failure by classifying the disk array device from level 1 to level 5. (ACM SIGMOD Conferance, Chicago, Illi
nois, June 1-3, 1988P109-P116).

The levels 1 to 5 for classifying the disk array devices proposed by David A. Patterson et al.
ID (Redundant Arrays of In)
expansive Disks) 1-5. The RAIDs 1 to 5 will be briefly described as follows. [RAID0] FIG. 16 shows a disk array device having no data redundancy, which is not included in the classification of David A. Patterson et al.
Called D0.

In the RAID 0 disk array device, as shown in data A to I, the disk array control device 10 distributes data to each of the disk devices 32-1 to 32-3 based on an input / output request from the host computer 18. Data redundancy in the event of a disk failure. [RAID1] FIG. 1 shows a RAID1 disk array device.
As shown in FIG. 7, a mirror disk device 32-2 stores copies A 'to D' of data A to D stored in the disk device 32-1. RAID 1 is widely used because it has low utilization efficiency of disk devices but has redundancy and can be realized by simple control. [RAID2] A RAID2 disk array device strips (divides) data in units of bits or bytes.
Reading / writing is performed in parallel to each disk device. The striped data is physically recorded in the same sector in all disk devices. A Hamming code generated from data is used as the error correction code. It has a disk device for recording a Hamming code in addition to the data disk device, identifies a failed disk device from the Hamming code, and restores the data.

[0006] By providing redundancy by the Hamming code in this way, correct data can be secured even if a disk device fails, but it has not been put to practical use due to poor utilization efficiency of the disk device. [RAID3] The disk array device of RAID3 has the configuration shown in FIG. That is, as shown in FIG. 19, for example, the data a, b, and c are divided into data a1 to a3, b1 to b3, c1 to c3 in units of bits or sectors, and a parity P1 is calculated from the data a1 to a3. b1
Ｂb3, a parity P2 is calculated from the data c1 to c3, and a parity P3 is calculated from the data c1 to c3.
1 to 32-4 are simultaneously accessed and written in parallel.

In RAID 3, data redundancy is maintained by parity. Further, the data writing time can be reduced by the parallel processing of the divided data. However, with one write or read access, all disk devices 3
2-1 to 32-4 parallel seek operations are required.
For this reason, it is effective when a large amount of data is handled continuously, but in the case of a transaction process in which a small amount of data is accessed at random, the high speed of data transfer cannot be utilized and the efficiency is reduced. [RAID4] As shown in FIG. 20, the RAID4 disk array device divides one data into sector units and writes the data into the same disk device. For example, the disk device 32
Looking at -1, data a is divided into sector data a1 to a4 and written. The parity is stored in a fixedly determined disk device 32-4. Here, data a1,
Parity P1 is calculated from b1 and c1, and data a2 and
Parity P2 is calculated from b2 and c2, and data a3
Parity P3 is calculated from b3 and c3, and data a4
Parity P4 is calculated from b4 and c4.

Data is read out from the disk devices 32-1 to 32-1.
32-3 can be read in parallel. Data a ~
The reading of data b is performed by taking the data a as an example.
2-1 access sectors 0 to 3 to access sector data a1
To a4 are sequentially read and synthesized. In data writing, since the new parity is calculated and written after reading the data and parity before writing, for one writing,
A total of four accesses are required. For example, when updating (rewriting) the sector data a1 of the disk device 32-1, the old data (a1) _old of the update location and the old parity (P1) _old of the corresponding disk device 32-4 are read, and the new data is read. An operation of writing a new parity (P1) _new that is consistent with the data (a1) _new is required in addition to the data writing for updating.

In addition, since writing always accesses the disk device 32-4 for parity, writing to a plurality of disk devices cannot be executed simultaneously. For example, even if the simultaneous writing of the data a1 of the disk device 32-1 and the simultaneous writing of the data b2 of the disk device 32-2 are performed, the parity P1, P2
Must be read and written after calculation, and therefore cannot be written at the same time.

Although the definition of RAID4 has been made in this way, there is little merit and there is little movement for practical use at present. [RAID5] The disk array device of RAID5 enables parallel reading and writing by not fixing the disk device for parity. That is, as shown in FIG.
The disk device on which parity is placed differs for each sector. Here, the parity P is calculated from the data a1, b1, and c1.
1 is calculated, and parity P is calculated from the data a2, b2, and d2.
2 is calculated, and parity P is calculated from data a3, c3, and d3.
3 is calculated, and parity P is calculated from data b4, c4, and d4.
4 has been calculated.

The parallel reading and writing is performed by, for example, the disk drive 3
2-1 data a1 of sector 0 and disk device 32-2
The data b2 of the sector 1 can be read and written simultaneously without duplication because the parities P1 and P2 are located in different disk devices 32-4 and 32-3. Note that the overhead that requires a total of four accesses at the time of writing is RAI
Same as D4.

As described above, RAID5 can access a plurality of disk devices asynchronously to execute read / write, and is therefore suitable for transaction processing in which a small amount of data is randomly accessed. Here, RAID3 to RAI
In the disk array device shown in D5, a combination of disk devices related to generation of redundant information is called a rank. For example, when there are k disk devices for recording data and m disk devices for recording redundant information related to data, (k + m) disk devices are collectively ranked. By connecting a plurality of ranks of disk arrays, the processing performance can be further improved.

[0013]

However, in a multi-rank disk array device, if the access path to each disk device is dual-ported, there is a problem that a deadlock occurs when performing a disk write process. Was. The reason for this will be described together with the write processing of RAID level 5 as follows. At RAID level 5, as shown in the following equation (1), the exclusive OR of data in each disk device is obtained and stored in the disk device as parity (error detection code). Data a (+) data b (+). . . = Parity P1 (1) where (+) indicates exclusive OR The storage locations of data and parity are such that parities P1 to P4 are distributed to the disk devices 32-1 to 32-4 as shown in FIG. This eliminates the concentration of accesses to one disk device due to the parity read / write operation (parity update operation). When reading data of RAID level 5, the data in the disk device is not rewritten, so that parity consistency is maintained. However, at the time of writing, the parity also needs to be changed in accordance with the data.

When updating old data in a certain disk device to rewrite the old data with new data, in order to obtain parity consistency, a calculation as shown in the following equation (2) is performed, and the disk device is updated by updating the parity. Parity consistency of the entire data can be maintained. Old data (+) old parity (+) new data = new parity (2) As can be seen from equation (2), in the data writing process,
It is necessary to read old data and old parity in the disk device first. In addition, since writing is performed in the same location as the location from which the data was read, when writing operation of the disk device is started, it takes time to perform an operation for one rotation of the disk without fail.

The disk device in which parity is further written is
In order to generate a new parity from the equation (2), it is necessary to wait until the old data is read and transferred from the disk device to which the data is written. FIG. 22 is a flowchart showing the processing operation of the RAID 5 level disk array device, and shows in parallel the processing operations of the data storage disk device and the parity storage disk device cooperating with the processing of the disk array control device.

Next, the reason why deadlock occurs will be described. First, a description will be given of a process in a case where two write requests (hereinafter, referred to as “transactions”) are made from a host in a disk array device having a single port configuration.
FIG. 23 shows a single-port disk array device in which the disk array 46 has a 4-rank configuration and the disk array control device 10 is only one.

In FIG. 23, 32-1 to 32-20 are disk devices, 34-1 to 34-5 are interfaces,
48-1 to 48-4 are ranks indicating array units.
The two disk devices 32-4 and 32- shown by the current hatched lines
17, the transaction 1 is issued to the disk array control device 10 as an update command to update the data D1 of the disk device 32-9 and to update the parity P1 of the disk device 32-7. -7
Of the transaction 2 as an update instruction to update the data D2 of the disk device 32-9 and the parity P2 of the disk device 32-9.
Are successively generated with little time.

Here, the disk devices 32-4, 32-17
And the interfaces 34-2 and 34-4 are in use, so that the instructions of the transactions 1 and 2 are queued until released. Disk devices 32-4, 32-1
7 are released, the interfaces 34-2, 34-4
In the transaction 1 via the disk device 32-7,
32-9 usage rights are obtained.

The D1 update instruction of the transaction 1 reads the old data D1 of the disk device 32-9, writes new data D1, and releases the disk device 32-9. The P1 update instruction of transaction 1 reads the old parity P1 from the disk device 32-7, and if the old data D1 is read in this state, the new parity P1 is generated from the equation (2). Parity P
After writing 1, the disk device 32-7 is opened.
Disk devices 32-7 and 32 by transaction 1
The access at -9 is performed simultaneously and in parallel.

After the transaction 1 is completed, the D2 update instruction and the P2 update instruction of the transaction 2 are executed in the same manner. Thus, in a single-port configuration,
Deadlock cannot occur because the transaction 2 generated later does not use the disk device first. FIG. 24 shows a disk array device having a dual port configuration, in which two disk array control devices 10-1 and 10 are provided.
-2, and interfaces 34-1 to 34-5,
Two access paths are formed by 36-1 to 36-5, and in principle, have twice the processing capacity as compared with a single port configuration.

In the case where accesses from the disk array controllers 10-1 and 10-2 overlap with respect to the same disk device, the one who has acquired the right to use first occupies the same disk device, and the other device obtains the disk until the access is completed. You will have to wait. The deadlock in the dual port configuration of FIG. 24 will be described as follows. Now, as shown in FIG.
The disk array control device 10-1 is a disk device 32-
4 and the disk control device 10-2 uses the disk device 32-17, and the disk array control device 10-
It is assumed that transaction 1 occurs in transaction 1 and transaction 2 occurs in disk array controller 10-2.

The previously generated transaction 1 includes a D1 update command for updating the data D1 of the disk device 32-9 and a P1 for updating the parity P1 of the disk device 32-7.
It has an update instruction and is queued. The transaction 2 that occurs later includes a D2 update instruction for updating the data D2 of the disk device 32-7 and the disk device 32-9.
And has a P2 update instruction to update the parity P2 of the same.

Here, the interface 3 from the disk array control device 10-1 to the disk device 32-7.
Since 4-2 is free, the right to use the disk device 32-7 is acquired by the P1 update instruction and the parity P is immediately obtained.
One can lead. Also, the disk device 32-
Since the interface 34-4 from the disk array control device 10-2 for No. 9 is also empty, the right to use the disk device 32-9 is acquired by the P2 update command, and the parity P2 can be read immediately.

When the parities P1 and P2 are read first, the transactions 1 and 2 respectively change the occupancy of the disk devices 32-7 and 32-9 to write the newly generated parity. The state is maintained, and the state shown in FIG. 25 is obtained. Next, transactions 1 and 2 are executed by using the D1 update instruction and the D2 update instruction to update the old data D
Attempt to generate a new parity by reading 1 and D2.
However, even if an attempt is made to access the disk device 32-9 by the D1 update instruction of the transaction 1, the disk device 32-9 is already occupied by the P2 update instruction of the transaction 2 and cannot be accessed. At the same time, if the user tries to access the disk device 32-7 with the D2 update command of the transaction 2,
It is already occupied by the P1 update instruction of transaction 1 and cannot be accessed similarly.

That is, transactions 1 and 2 occupy each disk device for parity update, and issue an acquisition request to use the disk device occupied by the other party and wait for each other to end. In this case, a deadlock occurs in which processing is stopped without opening either disk device. In other words, a deadlock occurs when a disk device that cannot be interrupted and can perform only one process at a time enters the cyclic standby state. SUMMARY OF THE INVENTION It is an object of the present invention to avoid a deadlock in a disk array configuration in which a plurality of ranks and dual ports are provided, and disk devices in the ranks can be individually accessed and parity is dispersed. A disk array device is provided.

Another object of the present invention is to provide a disk array apparatus which determines the possibility of establishment of a specific deadlock from the contents of successive transactions and performs processing. Another object of the present invention is to provide a disk array apparatus which sets a processing procedure of a transaction in which deadlock cannot occur.

Another object of the present invention is to provide a disk array device in which, when the possibility of deadlock establishment is determined, the processing procedure of a subsequent transaction is changed so as to avoid deadlock. Another object of the present invention is to provide a disk array device that processes transactions in order to avoid deadlock.

[0028]

FIG. 1 is a diagram illustrating the principle of the present invention. First, the present invention provides a disk having a disk array 46 provided with a plurality of sets of a plurality of disk devices 32 storing a plurality of data processed in parallel and a parity as redundant information of the data as one array unit (rank). Targets array devices.

The parity stored in each disk device 32 of the disk array 24 is a data storage format equivalent to RAID 5 level and a parity generation format stored in a different disk device each time the data storage position changes. Further, at least two disk array control means 10-
1, 10-2 are provided, and a dual port configuration having at least two access paths for individually accessing the disk device 32 from different paths to read / write data or parity is provided.

The RA which has a plurality of ranks and dual ports, and further distributes the parity storage disk device.
According to the present invention, a disk array device corresponding to the ID5 level is used as the disk array control means 10-1 and 10-2.
Thus, a deadlock suppressing means 34 for suppressing a deadlock when two or more processing requests for simultaneously accessing two disk devices belonging to the same array unit are present before and after is provided. I do.

Here, a deadlock occurs when the disk array processing means 10-1 and 10-2 have transactions 1 and 2 generated before and after, and the transaction 1 updates the data D1 of one disk device and the transaction 2 The parity P2 of the other disk is updated, and the transition 2 is the parity P2 of one disk device.
This may occur when the condition for updating transaction 1 or updating the data D2 of the other disk device is satisfied.

That is, if both the transactions 1 and 2 update the parity data before the deadlock occurrence condition is established, a deadlock occurs. Therefore, deadlock can be avoided by not starting with the parity update processing or by opening the disk at the end of parity read even if parity is updated first.
The processing performed by the deadlock suppressing means 34 is as follows:
There are five. [Case 1] Old data D1, D2 and old parity P
Reading of 1, P2 and writing of new parity are performed separately. That is, when the reading of the old parities P1 and P2 is completed, the disk device for the parity is released, and the data D1 and D2 can be updated by the other transaction. When the data D1 and D2 have been updated, the old data D
Since D1,1 and D2 have already been read, new parity P1, P2 may be generated, the disk device may be acquired again, and new parity may be written. [Case 2] A procedure in which the data storage disk device and the parity storage disk device are not accessed simultaneously in parallel, and the parity storage disk cannot be accessed unless the data storage disk device is always accessed. Is determined. [Case 3] When reading of the old parity P1 is completed in the preceding transaction 1, the access progress of the data disk device is referred to, and the disk device is occupied and released according to the following cases.

When the data storage disk device is accessing, for example, seeking the old data D1 and D2, transferring the buffer, transferring the old data to the buffer, and the like, the old data is read by updating the data after a while, and the new parity is read. Since the disk device can be released by generating and writing, the disk device is kept occupied until the parity update is completed.

If the data disk device cannot be acquired and the device is in the queue, the condition for establishing a deadlock is determined. If the condition is satisfied, the transaction 2 generated later is currently occupied. The disk device for parity is released, and the data update by the transaction 1 that has occurred earlier is processed first. [Case 4] The deadlock establishment condition is determined before accessing the parity disk unit in the succeeding transaction 2, and if it is satisfied, the access progress status of the data disk unit of the preceding transaction 1 is referred to, and Until the access to the parity disk device is started, access to the parity disk device by the subsequent transaction 1 is prohibited. [Case 5] Before assigning all transactions to the disk array control means 10-1 and 10-2, the transactions are stored in the order of occurrence in the first-in first-out queue, and sequentially stored in the disk array control means 10-1 and 10-2. Deliver and execute. Therefore, a request for access to the disk device of a transaction generated later is not processed earlier than a request for access to the disk device of a previously generated transaction, and deadlock cannot occur.

[0035]

According to the disk array control apparatus of the present invention having such a configuration, a so-called RAID5 which can individually read and write each disk device of the disk array and changes the parity every time the position of the stored data is different. Prevents deadlocks by predicting deadlocks that occur under specific conditions and establishing conditions that do not allow deadlocks when a dual-port configuration with multiple ranks is configured for a level equivalent configuration it can.

Even if a deadlock occurs, the disk can be forcibly released to avoid the deadlock or quickly recover from the deadlock state.

[0037]

【Example】

<Table of Contents> 1. Hardware configuration of the present invention 2. Processing function of the present invention 3. First Embodiment of Deadlock Suppression Processing of the Present Invention 4. Second Embodiment of Deadlock Suppression Processing of the Present Invention 5. Third Embodiment of Deadlock Suppression Processing According to the Present Invention 6. Fourth embodiment of deadlock suppression processing according to the present invention Fifth Embodiment of Deadlock Suppression Processing of the Present Invention FIG. 2 is a block diagram of an embodiment showing a hardware configuration of the disk array device of the present invention.

In FIG. 2, the disk array controller 1
0 is provided with two systems of disk array control units. That is, two MPUs 12-1 and 12-2 are provided as control means in the disk array control device 10.
Taking the PU 12-1 as an example, a host interface 16-1 for exchanging with the host computer 18 on the internal bus 14-1, a ROM 2 fixedly storing programs and the like.
0-1, R used as control storage or buffer
AM 22-1, cache memory 26-1, connected via cache control unit 24-1, data transfer buffer 2
8-1.

Similarly, the MPU 12-2 has a host interface 16-2, a ROM 20-2, a RAM 22-2,
Cache memory 2 via cache control unit 24-2
6-2 and the data transfer buffer 28-2. On the other hand, in this embodiment, the disk array 46 is composed of 20 disk devices 32-1 to 32-20, and five disks constitute one rank, and the rank 48-1.
It has 4 ranks of ~ 48-4. Also, the disk device 32
-1 to 32-20 have a dual port configuration accessible from two different systems.

Each disk device 32 of the disk array 46
Reference numerals 3-1 to 32-20 denote device adapters 30-11 to 30-15 and 30-21 to 30-25 functioning as interface control units provided in the disk array control device 10.
Through the internal bus 14- of the MPUs 12-1 and 12-2.
1, 14-2. Further, a dual port RAM 56 functioning as a communication control storage is connected between the internal buses 14-1 and 14-2. 2. FIG. 3 is a functional block diagram showing the processing functions of the disk array device according to the present invention.

In FIG. 3, first, two disk array controllers 10-1 and 10-2 are separately provided so that control signals and data can be exchanged with the host computer 18, respectively. The disk array control device 10-1 is provided with a control unit 38-1 implemented by MPU program control. The control unit 38-1 includes a control storage 40-1, a data processing unit 44-1 and a parity processing unit. The function as 45-1 is provided.

The data processing unit 44-1 determines the logic I based on the result of decoding the write command from the host computer 18.
A process of rewriting data stored in a specific disk device specified by D and a sector address is executed. Further, the parity processing unit 45-1 performs a parity writing process in accordance with data rewriting by the data processing unit 44-1. Here, since the data storage format and parity generation format in the disk array device of the present invention are equivalent to the RAID5 level shown in FIG. 16, the data processing unit 44-1 needs to use the old format to generate a new parity. Data read processing is required, and even the parity processing unit 45-1 requires old data read processing.

More specifically, the control unit 38-
When the host computer 18 receives a host command as a processing request, the control unit 38-1 creates a data update instruction and a parity update instruction and issues them to the specified lower-level disk device. Here, the data update instruction is data update instruction = (data read instruction) + (data write instruction), and the parity update instruction is parity update instruction = (parity read instruction) + (parity generation instruction) + (parity) Write instruction). The data processing unit 44-1 issues a data read command in the data update command as one package command to the target disk device to access data read and data write.

The parity processing unit 45 first issues a write command to the target disk device, and when the old parity can be read by this, confirms that the data processing unit 44-1 has read the old data and generates the parity. Part 50-1
Is requested to generate a new parity, and when a new parity is generated, a parity write command is issued to write the new parity.

The parity generation unit 50-1 performs exclusive control of the old data read from the disk to be written, the old parity read from the same sector position of another disk belonging to the same rank, and the new data given from the host computer 18. Generate a new parity from the logical sum. Further, the disk control device 10-1 is provided with a buffer unit 52-1 and an interface control unit 54-1. The buffer unit 52-1 stores new data, old data, and old parity when writing data. The interface control unit 54-1 is connected to each of the disk devices 32-1 to 32-32 in the disk array.
n is accessed.

Such a disk array controller 10-
Configuration 1 is the same for the disk array control device 10-2. Further, the disk array controller 10-
In order to enable control information to be exchanged between 1 and 10-2, a communication control storage 56 using a dual port RAM or the like is provided.
Necessary information can be exchanged between each other by reading and writing by 8-1 and 38-2.

The disk array 46 in the functional block diagram of FIG. 3 includes disk devices 32-1, 32-2,.
Only one rank of 32-n is shown, and among these, transactions 1 and 2 as two data write processing requests are sent to the disk devices 32-1 and 32-2 by the disk array control devices 10-1 and 10-. Data D corresponding to the deadlock occurrence condition that occurs when each of the two occurs
1 and D2 and the storage states of the parities P1 and P2.

Disk array control devices 10-1 and 10-
2 are provided with deadlock management tables 42-1 and 42-2 in the control memories 40-1 and 40-2.
FIG. 4 is an explanatory diagram of the deadlock management table used in the present invention, in which information on the transaction number, the parity storage disk, and the data storage disk is written. Further, the contents of the parity storage disk and the data storage disk are divided into two parts: a disk number and access information.

FIG. 5 shows an operation state immediately before a deadlock occurs in the disk array device of the present invention.
This is the same state as the description of the deadlock in FIG. 19, and is different in that information can be exchanged between the disk array control devices 10-1 and 10-2. That is, in the state of FIG.
Disks 32-4 and 32-17 are in use by two transactions at -1 and 10-2. If the disks 32-4 and 32-17 are used for parity and new parity is being written, the data disks are, for example, 32-1 and 32-20. I do.

In this state, transaction 1 and transaction 2 occur.
This is the data write of 2-9. Therefore, the D1 update instruction for the disk 32-9 and the P1 update instruction for the disk device 32-7 are queued by the transaction 1. In the disk array control device 10-2, a D2 update command for the disk device 32-7 and a P2 update command for the disk device 32-9 by the transaction 2 generated with little time to the transaction 1 are queued.

In the operation state as shown in FIG. 5, the deadlock management table used in the present invention has the registered contents as shown in FIG. Therefore, by referring to the deadlock management table, it is possible to determine whether the deadlock occurrence condition is satisfied or to know the disk access status. 3. First Embodiment of Deadlock Suppression Processing of the Present Invention FIG. 6 is a time chart showing a first embodiment of deadlock suppression in the present invention.

First, a deadlock occurs in the disk drive 32-
1, D2-2 and parity P1,
For the storage state of P2, the continuous transaction 1,
2 occurs in the disk array controllers 10-1 and 10-2 and occurs when the parity P1 and P2 are updated. Therefore, in the first embodiment, as shown in FIG.
1 and P2, the disk devices 32-1 and 32-2 are opened upon completion of the parity read, and the data D
1 and D2 are accessed (read / write).

When the data D1 and D2 have been updated by reading and writing and the disk devices 32-1 and 32-2 have been released, the disk device is acquired again and the calculated new parity is written. FIG. 7 is a flowchart showing the processing of the first embodiment for suppressing deadlock in the present invention. Note that the processing operation of the disk array control device shown in FIG.
2 will be executed in parallel.

First, in step S101, the disk array controller receives a write command from the host. For example, taking the disk array controller 10-1 as an example, a write command for the data D1 area of the disk device 32-2. To receive. Subsequently, in step S102, a data update instruction is issued to the disk device 32-2 specified by the logical ID of the write command. Specifically, the data update instruction issuance process first issues an access request to the disk device 32-2, and if it is a busy response, queues the access request.
If a ready response is obtained, the access right is acquired, the connection is established via the interface, and then a data update command is transmitted.

In step S103, the disk device 32-2 for data writing and the data write sector are specified, and the disk 32-1 storing the parity data is specified at this sector position. Disk device 32
Issue a parity update instruction for -1. As with the data update instruction, an access request is first issued to the disk device 32-1. If the response is a busy response, the access request is queued. When a ready response is obtained, the access right is acquired. To transmit the parity update command.

The processing of the data storage disk device activated by the data update command is performed in steps S201 to S206.
It becomes as shown in. That is, a seek for positioning the head at the head of the designated write block is executed in step S201, and when the read is completed due to the seek completion in step S202, the old data is read out in step S203 and stored in the buffer unit on the disk array controller side. Read and transfer.

Subsequently, after waiting one rotation in step S204,
Since the data becomes writable in step S205, the new data is written to the disk in step S206, a series of processing is completed, and the disk device to which the data has been written is released. On the other hand, in the parity storage disk device started by the parity update command, seek processing is started in step S301, and if the head can be positioned at the head of the read block in step S302, it becomes readable. The data is read out and read out and transferred to the buffer on the disk array controller side. Subsequently, after one rotation of the disk in step S304, it is checked in step S305 whether a new parity is generated.

On the disk array controller side, when reading of the old data and the old parity is completed, a new parity is generated and transferred.
When the disk is rotated in step S304 and the head is positioned at the head of the block and becomes writable, parity data is written to the disk in step S307, and the disk device for parity storage is opened.

The basic processing operation of such a data storage disk device and a parity storage disk device is the same as that of the prior art. However, in the first embodiment, parity storage is performed by the processing of the disk array controller. After the parity data is read out and transferred to the buffer in step S303 of the disk device for use, the disk is forcibly released.

For this reason, in step S103 of the disk array control device, it is monitored whether or not the reading of the old parity is completed after the issuance of the parity update command. Proceeding to S104, the disk device on the parity side is forcibly released. When the parity side disk device is forcibly released, the data update instruction of transaction 2 is queued on the disk array controller 10-2 side simultaneously and in parallel at this time. , And the access process is performed by the data update instruction of the transaction 2.

When the data update processing on the transaction 2 side is completed, the disk device is released. Therefore, the disk array controller monitors in step S107 whether the disk device on the parity side can be accessed. Then, the access on the parity side is restarted in step S108, and the processing from step S304 in the parity disk device is restarted.

During this time, in the disk array controller, the completion of reading of data and parity is determined in step S105. Therefore, parity generation processing is performed in step S106, and disk access on the parity side is performed in step S108. Upon resumption, a new parity is sent from the disk array controller. When writing becomes possible from step S305 to step S306,
The new parity can be written immediately in step S307.

The disk array control device proceeds to step S10
After restarting the access to the disk device on the parity side in step S8, the completion of writing of data and parity is monitored in step S109. When both writings are completed, a write completion notification is sent to the host computer to execute a series of processing. finish. 4. Second Embodiment of Deadlock Suppression Process of the Present Invention FIG. 8 is a time chart showing a second embodiment of the deadlock suppression process of the present invention. In the second embodiment, regardless of whether a deadlock occurrence condition is satisfied or not, when a transaction occurs, access to the data storage disk device is started before the parity storage disk device is accessed. It is characterized in that access is started. For example, when the transaction 1 occurs, the access to the data storage disk device 32-1 is started before the access to the parity storage disk device 32-1 is started. The same applies to transaction 2.

More specifically, taking transaction 1 as an example, processing is performed so that access to the parity storage disk device 32-1 is not permitted until it is determined that access to the data storage disk device 32-2 is possible. I do. FIG. 9 is a flowchart showing a deadlock suppressing process according to a second embodiment of the present invention.

In the disk array control device, upon receiving the host command in step S101, it is determined in step S102 whether or not access to the data-side disk device is enabled. When the disk device on the data side obtains an access point by obtaining a ready response, the process proceeds to step S103, and issues a data update command to the data storage disk device. Also, the fact that access has been started is written in the corresponding access state of the deadlock management table shown in FIG.

Next, in step S104, it is checked whether or not the disk device on the parity side is accessible. If the access right is acquired by a ready response, the flow advances to step S105 to issue a parity update instruction. Subsequent processing waits for completion of reading of old data and old parity in step S106, generates parity in step S107, finally confirms writing of data and parity, and ends a series of processing. The processing in the data storage disk device and the parity storage disk device is the same as that of the first embodiment except that the disk is not released halfway, and when data writing and parity writing are completed, FIG. Rewrite the access status of the indicated deadlock management table to “written”. 5. Third Embodiment of Deadlock Suppression Process According to the Present Invention FIG. 10 is a flowchart showing a third embodiment of the deadlock suppression process according to the present invention. In the third embodiment, when the parity data P1 is read from the disk device 32-1 in the previously generated transaction 1, the progress of access to the data storage disk device 32-2 in the same transaction 1 is referred to. If it is in use, wait a little while and the data storage disk device 3
Since the old data is read from 2-2 and the new parity can be calculated, the occupation state of the disk device 32-1 for storing the parity is maintained, and the disk device 32-1 is opened after continuing to the parity write.

On the other hand, as shown in the time chart of FIG.
When the disk device 32-2 is occupied by the subsequent transaction 2 in the read state of the parity P1 of 2-1 and the update instruction of the data D1 is in the queue, the condition is that the deadlock is established. Then, the occupation of the disk device 32-2 by the subsequent transaction 2 is forcibly released, and the access to update the data of the preceding transaction 1 is terminated first.

FIG. 12 is a flowchart showing a third embodiment of the deadlock suppressing process according to the present invention. FIG.
Is a process on the side that executes the previously generated transaction 1. First, when a command is received in step S101, a data update instruction is issued in step 102 to access the data storage disk device for data update.

Next, in step S103, the transaction number and the disk number of the parity disk device are notified and registered in the deadlock management table of the disk array control device of the self and the other party processing the subsequent transaction 2. I do. Based on the notification of the transaction number and the disk number, it is possible to determine from each deadlock management table whether the deadlock occurrence condition is satisfied.

Next, in step S104, a parity update command is issued to the parity storage disk device. Subsequently, the completion of parity reading is monitored in step S105. When the old parity is read and transferred to the buffer, the process proceeds to step S106, and the access progress of the data storage disk device is referred to by the deadlock management table. .

At this time, based on the data update command issued in step S102, the data storage disk device is accessing during the seek of step S201, the readable state of step S202, and the access of the read data in step S203 during buffer transfer. The old data can be obtained on the disk array controller side only by waiting a short time.

Then, the process proceeds to step S110 while maintaining the control state of the parity storage disk device, and confirms the reading of the old data and the old parity.
The parity is generated in 1 and sent to the parity storage disk device, and the parity is written in the processing of steps S304 to S307, and the parity storage disk device is opened.

Therefore, when the parity storage disk device is released, the access right of the disk device that executes the data update instruction of the subsequent transaction in the queue is acquired, and no deadlock occurs. On the other hand, if it is assumed in step S107 that the data storage disk device of transaction 1 has been in the queue due to the occupation of the disk device by executing the parity update command of the subsequent transaction 2, the deadlock management table is referred to in step S108. It is determined whether a deadlock condition is satisfied.

At this time, if the deadlock condition is satisfied, the flow advances to step S109 to execute the data update command queued in transaction 1 for the parity storage disk device occupied by the subsequent transaction 2. For this reason, once open. For this reason, the data update command issued in step S102 that has been in the queue acquires the access right in response to the release of the disk on the transaction 2 side, and performs data update processing by the data storage disk device. When the data update processing of the transaction 1 is completed and the disk device is released, the parity update of the transaction 2 forcibly released in step S109 is resumed, and the parity generation and the parity writing of the transaction 2 are performed. . 6. Fourth Embodiment of Deadlock Suppression Processing According to the Present Invention FIG. 13 is a time chart showing a fourth embodiment of the deadlock suppression processing according to the present invention. In the fourth embodiment, it is determined whether or not a deadlock condition is satisfied before accessing the parity storage disk device 32-2 in a transaction 2 generated later.
The parity access of the transaction 2 is not permitted until the data access of the transaction 1 is started by referring to the access progress state of the data storage disk device 32-1 of the transaction 1 generated earlier.

That is, the third embodiment shown in FIGS. 110 to 12 is the processing of the preceding transaction 1 in the disk array controller, whereas the fourth embodiment shown in FIG. The processing is performed by the array controller. FIG. 14 is a flowchart showing a fourth embodiment of the deadlock suppression processing according to the present invention. First, in step S101, a command from the host computer is received, and when a subsequent transaction 2 occurs, a data update instruction is issued in step S102. Then, access to the data storage disk device is started.

Next, in step S103, it is determined whether or not the deadlock condition is satisfied by referring to the deadlock management table. If the deadlock condition is satisfied, the data access of the preceding transaction 1 is advanced in step S104. See the situation. At this time, if the data update instruction is queued without starting the data access of the preceding transaction 1, the process waits until the data access is started.

When data access is started in the preceding transaction 1, parity access in the following transaction 2 is permitted, and a parity update command is issued in step S106 to start the parity update process by the parity storage disk device. After that, the read of the old data and the old parity is confirmed in step S107, and the parity is generated in step S108.
When the writing of data and parity is confirmed in step 9, the end of writing and the disk number are written in the deadlock management table in step S110, and a series of processing ends. 7. Fifth Embodiment of Deadlock Suppression Process of the Present Invention FIG. 15 shows a FIFO command queue in a disk management table used in a fifth embodiment of the deadlock avoidance process of the present invention.

That is, in the fifth embodiment of the present invention, before all the transaction requests are assigned to the disk devices, once, for each access request to all the disk devices, first-in and first-out operations are performed as shown in FIG. And a command queue for the FIFO.
In FIG. 15, the FIFO queue stores a data update instruction and a parity update instruction for realizing the requests of transaction 1 and transaction 2, and the data update instruction includes two access instructions of a data read instruction and a data write instruction. Therefore, these are stored in order. Further, since the parity update instruction is three access instructions of a parity read instruction, a parity generation instruction, and a parity write instruction, three access instructions are also arranged in order.

Each access instruction put in the FIFO command queue shown in FIG. 15 is sequentially taken out from the bottom to access the corresponding disk device, and the data update instruction and parity for the same disk device in transactions 1 and 2 are obtained. Since update instructions do not overlap, deadlock can be reliably prevented. In the above embodiment, one disk is constituted by five disk devices,
In the above, a four-port configuration is used, and a dual port configuration using two access buses for each disk device is taken as an example. However, the present invention is not limited to this.
The number of disk devices, the number of ranks, and the number of access ports per rank can be determined as needed.

[0080]

As described above, according to the present invention, in a disk array device having a dual-port configuration with a plurality of ranks, a deadlock that occurs when parity update and data update from different transactions overlap on the same disk. The input / output processing can be appropriately performed while preventing or avoiding the problem, and the throughput of the disk array device operating in parallel by the dual port can be improved.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating the principle of the present invention.

FIG. 2 is a configuration diagram of an embodiment showing a hardware configuration of the present invention.

FIG. 3 is a functional block diagram showing the processing functions of FIG. 2;

FIG. 4 is an explanatory diagram of a deadlock management table used in the present invention.

FIG. 5 is an explanatory diagram showing a generation state of a transaction in which a deadlock occurs.

FIG. 6 is a time chart showing a first embodiment of a deadlock suppressing process according to the present invention;

FIG. 7 is a flowchart showing a first embodiment of a deadlock suppression process according to the present invention;

FIG. 8 is a time chart showing a second embodiment of the deadlock suppression processing of the present invention.

FIG. 9 is a flowchart illustrating a deadlock suppression process according to a second embodiment of the present invention;

FIG. 10 is a time chart showing a third embodiment of the deadlock suppression processing of the present invention.

FIG. 11 is a time chart showing another deadlock suppression process in the third embodiment.

FIG. 12 is a flowchart showing a deadlock suppressing process according to a third embodiment of the present invention;

FIG. 13 is a time chart showing a fourth embodiment of the deadlock suppression processing of the present invention.

FIG. 14 is a flowchart illustrating a deadlock suppressing process according to a fourth embodiment of the present invention;

FIG. 15 is an explanatory diagram of control storage used in a fifth embodiment of the deadlock suppression processing of the present invention.

FIG. 16 is an explanatory diagram of a disk array device using RAID0.

FIG. 17 is an explanatory diagram of a disk array device using RAID1.

FIG. 18 is an explanatory diagram of a disk array device using RAID3.

FIG. 19 is an explanatory diagram of data division in RAID3.

FIG. 20 is an explanatory diagram of a disk array device using RAID4.

FIG. 21 is an explanatory diagram of a disk array device using RAID5.

FIG. 22 is a flowchart showing data write processing of a disk array device by RAID5.

FIG. 23 is an explanatory diagram showing that deadlock does not occur in a single-port configuration.

FIG. 24 is an explanatory diagram showing a state immediately before deadlock occurs in a dual port configuration;

FIG. 25 is an explanatory diagram of a state where deadlock has occurred in a dual-port configuration.

[Explanation of symbols]

10-1, 10-2: Disk array controller 12-1, 12-2: MPU 14-1, 14-2: Internal bus 15: System bus 16-1, 16-2: Host interface 18: Host computer ( 20-1, 20-2: ROM 22-1, 22-2: volatile memory (RAM) 24-1, 24-2: cache control unit 26-1, 26-2: cache memory 28-1 , 28-2: Data transfer buffer 30-11 to 30-15, 30-21 to 30-25: Device adapter 32-1 to 32-n: Disk device 34-1 to 34-5, 36-1 to 36- 5: interface 38-1, 38-2: control unit 40-1, 40-2: control storage 42-1, 42-2: deadlock management table 44-1, 44-2: data processing unit 45- , 45-2: Parity processing unit 46: Disk array 48-1 to 48-4: Rank 50-1, 50-2: Parity generation unit 52-1, 52-2: Buffer unit 54-1, 54-2: Interface control unit 56: Communication control storage 60: Deadlock suppression unit

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-59-5331 (JP, A) JP-A-1-140326 (JP, A) JP-A-4-278641 (JP, A) JP-A-4- 312146 (JP, A) (58) Field surveyed (Int. Cl. ⁶ , DB name) G06F 3/06 301 G06F 3/06 540 G06F 11/30 305

Claims (8)

(57) [Claims] A plurality of disk devices storing a plurality of data to be processed in parallel and redundant information of the data;
A plurality of sets as a single array unit, and redundant information stored in each disk device of the disk array (46) is stored in a different disk device every time a data storage position is changed, and a different path is used. At least two disk array control means (10) for individually accessing disk devices to read and write data or redundant information.
-1 and 10-2), and deadlock when two processing requests for simultaneously accessing two disk devices belonging to the same array unit by the disk array control means (10-1 and 10-2) exists. A disk array device comprising: a deadlock suppressing means (34) for suppressing. 2. The disk array device according to claim 1, wherein said deadlock suppressing means (34) is provided in a state where said two disk array processing means (10-1, 10-2) are generated before and after. The first processing request updates data (D1) at a specific position of the first disk device, and the second processing request updates redundant information (P2) at a different position on the first disk. And the first processing request is the second
When the redundant information (P1) at the specific position of the disk device is updated and the data (D2) is updated to a different position on the second disk device by the second processing request, it is determined that there is a possibility of deadlock. A disk array device characterized by the above-mentioned. 3. The disk array device according to claim 1, wherein said deadlock suppressing means (34) once releases the parity disk device after reading the redundant information (P1, P2) based on the processing request. , A data update by reading and writing data (D1, D2) before generating and writing redundant information. 4. The disk array device according to claim 1, wherein said deadlock suppression means (34) starts updating data (D1, D2) based on the first and second processing requests first. After the start of data update, the redundant information (P1, P
A disk array device which permits the start of the update in 2). 5. The disk array device according to claim 1, wherein said deadlock suppressing means (34) is operative when reading of redundant information (P1) based on a previously generated first processing request is completed. The update status of the data (D1) based on one processing request is referred to, and during the data update, the occupation state of the disk device is maintained until the end. When the possibility of deadlock is determined from the relationship, the disk device in use is released to update the redundant information (P2) based on the second processing request generated later, and the first processing request generated earlier The data (D1) of the disk array device in advance. 6. The disk array device according to claim 1, wherein said deadlock suppressing means (34) performs a deadlock before starting access to a disk device for updating redundant information by a second processing request generated later. When the possibility of locking is determined, the access progress state of the disk device for which data is updated by the first processing request generated earlier is referred to, and the redundant information by the second processing request generated later until the access is started. A disk array device for inhibiting access to a disk device to be updated. 7. The disk array device according to claim 1, wherein said deadlock suppression means (34) sends a processing request to said first and second disk array control means (10-1,
Before distribution to 10-2), the queues are arranged in a first-in first-out queue, and the first or second disk array control means (10-1, 10-2) is arranged in order from the processing request generated earlier.
A disk array device characterized by being executed by: 8. The disk array device according to claim 1, wherein said first and second disk array control means (10-
1, 10-2) uses parity data as redundant information, and generates new parity data from exclusive OR of old data before update, old parity, and new data to be updated. apparatus.