JPH07261945A - Disk array device and disk array dividing method - Google Patents

Abstract

PURPOSE:To lower the probability of disappearance of data and to improve the performance against the faults by successively recovering the data of the faulty drives in order of lower parity group levels. CONSTITUTION:A logical group 10 mixedly includes a parity group which constructs a level 5 with 5 pieces of data and a parity (5D+1P), a parity group which constructs a level 5 with 3 pieces of data and a parity (3D+1P) and a parity group obtained by duplication. These parity group levels are sorted in order of higher reliability and performance of them. For instance, the first drive 12 has a fault and the data stored in this drive are recovered. Under such conditions, the parity groups are successively processed for their faults in order of lower levels (higher probability of disappearance of data when the second drive has a fault).

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 suitable for being applied to a computer system or the like, and more particularly to a disk array device and a disk array partitioning method that enable high performance input / output operation.

[0002]

2. Description of the Related Art In a current computer system, data required by a high-order side such as a CPU is stored in a secondary storage device, and the data is stored in the secondary storage device as needed by the CPU. -Writing and reading data. A non-volatile storage medium is generally used as the secondary storage device, and typical examples thereof include a magnetic disk device (hereinafter referred to as a drive) and an optical disk.

In recent years, as computerization has advanced, there has been a demand for higher performance of this type of secondary storage device in computer systems. As one solution, a disk array composed of a large number of drives each having a relatively small capacity is considered.

As a known document, "D. Patterson, G. Gibs
on, and RHKartz; A Case for Redundant Arrays of In
expensive Disks (RAID), in ACM SIGMOD Conference, Chi
cago, IL, (June 1988) ”. In this document, a disk array (level 1) for duplicating data, a disk array (level 3) for dividing data and processing in parallel, and a disk array (level 4, 5) for separately handling data separately. ), The results of examination of its performance and reliability have been reported. The method currently described in this paper is considered to be the most common disk array.

First, a level 3 disk array will be briefly described. In the level 3 disk array, one piece of data given from the upper level is divided and divided among a plurality of drives for storage. To read that data,
Divided data pieces are collected from multiple drives, combined, and transferred to a higher level. Therefore, at level 3, parallel processing can be performed by a plurality of drives, and the transfer speed can be improved.

Next, a disk array (level 5) in which individual data is distributed without dividing the data and handled independently will be described. Level 4 is such that the parity distributed in each drive forming the disk array in level 5 is stored in one drive that stores only parity.

In the level 4 and level 5 disk arrays, individual data is handled independently without being divided, and is distributed and stored in a large number of relatively small capacity drives. In a secondary storage device of a general-purpose large-scale computer system that is generally used at present, since the capacity per drive is large, the drive is used for another read / write request, and the drive cannot be used and is kept waiting. Many things happened. In this level 5 (or 4) disk array,
The large-capacity drive used in the secondary storage device of a general-purpose large-scale computer system is composed of a large number of relatively small-capacity drives to store data in a distributed manner, which increases read / write requests. However, it becomes possible to perform distributed processing by a plurality of drives in the disk array, and the waiting time for read / write requests is reduced.

Next, the parity in the disk array will be described. Since the disk array comprises a conventional large-capacity drive with a large number of relatively small-capacity drives, the number of parts increases and the probability of failure increases. For this reason, parity is prepared in the disk array.

FIG. 5 shows a method of creating parity in level 5 (same in level 4), and FIG. 3 shows a storage state of these data and parity in level 5 of the conventional disk array.

As shown in FIG. 5, parity (Parit)
y) is data # 1 (Data # 1) to data # 5
It is created by taking the exclusive OR for each corresponding bit between each data of (Data # 5). The parity created in this way is stored in a drive other than the drive in which the data involved in creating the parity is stored.

Specifically, as shown in FIG. 3, independent data # 1 to data # 5 are stored in any one of the drives # 1 to # 6 and are created from these data. The parity is stored in a drive other than the drive in which the data is stored.

Since FIG. 3 shows a level 5 disk array, the parity is 6 drives # 1 to # 6.
It is distributed and stored in. At level 4, since the parity is stored in one drive, the parity is collectively stored in drive # 6, for example.

In FIG. 3, the drives # 1 to # 6 constituting the disk array are called a logical group. In addition, a set of data in which a parity is created (data # 1 to # 5 in FIG. 5) and a parity created from these data in the logical group is called a parity group. In FIG. 3, six drives # 1 to # 1
A logical group is constituted by # 6, and all parity groups in this logical group are constituted by 5 data and 1 parity.

When a failure occurs in any one of the drives # 1 to # 6 forming the logical group, each data in the failed drive must be restored using the data of a normal drive. You can That is, for each parity group to which each piece of data in the failed drive belongs, if the exclusive OR of the data in the normal drive and the parity is taken, each piece of data in the failed drive can be restored. You can

On the other hand, Japanese Patent Laid-Open No. 5-257611 discloses a method of dividing a logical group into a plurality of areas (partitions) and setting different RAID levels for the respective areas thus divided. In this method, for example, when a logical group consists of 6 drives,
Each area spans 6 drives or is divided into 6 areas for each drive. Then, by forming a logical unit using such an area, a plurality of logical units are formed on a single disk drive set.

[0016]

The conventional disk array has the following two problems.

First, the first problem is availability. In the disk array, even if a failure occurs in any one drive in a certain logical group, it is possible to recover by the parity, and therefore data loss does not occur.
However, if a failure occurs in any of the remaining drives in the same logical group before recovery of the first failed drive, recovery by parity is impossible and data will be lost.

For this reason, if the number of drives forming the logical group (the number of data forming the parity group) is large, the probability of occurrence of the second failure increases, and the probability of data loss increases. Therefore, in order to avoid such data loss, if a drive failure occurs, no matter how important the processing is, it will be stopped.
It is necessary to recover data as soon as possible.

In Japanese Patent Application No. 3-94728, two parities are provided in a parity group and a recovery process is performed depending on whether one drive fails or two drives fail. A method of changing the way is disclosed. This is because in the logical group, if the first drive fails, the data will not be lost even if the other drive is broken, so recovery processing is performed relatively slowly, and two drives fail. If one more
If a failure occurs in one drive, data will be lost, so recovery processing is performed immediately.

With this method, availability is improved because one drive failure does not require immediate recovery.
However, since it is necessary to uniformly prepare two pieces of parity for all the parity groups forming the logical group, the capacity used for the parity is further reduced to one.
The drive is required, which increases the cost.

In the future, the file server, which is a major application destination of the disk array, will become larger in size, and a large number of users will use one file server. In such an environment, since the file server is required to have high availability, it is difficult to immediately stop the disk array for a long time to recover data even if a drive fails. Furthermore, in the file server, it is necessary to reduce the cost paid for such availability as much as possible.

The second problem is the deterioration of performance at the time of failure. As described above, in the disk array, when one parity is prepared in the parity group, if one drive in the logical group fails, the data in the drive in which the failure occurs will be the parity group. It is possible to recover by the data and parity in the remaining normal drive in units. When this function is used, any drive in the logical group fails, and CP for the data in the failed drive
When a read or write request is issued from U,
It can be accepted by using the function of the recovery process.

Specifically, when the CPU issues a request to read the data of the failed drive, the data in the failed drive is restored from all the data and parity of the normal drives in the logical group, Transfer to CPU. When the CPU issues a write request to the failed drive, parity is created from all the data and write data of the normal drive in the parity group, as in the case of reading, and the drive in which the parity is stored. The parity is updated by storing in.

As described above, when a read or write request is issued to a drive in which a failure has occurred before the recovery processing is performed, the read request is issued to all the drives in the logical group, resulting in a large decrease in performance. To do. In particular, as the number of drives forming the logical group (the number of data forming the parity group) increases, the number of read requests issued increases, resulting in a large decrease in performance.

On the other hand, according to the method of the above-mentioned Japanese Patent Laid-Open No. 5-257611, it is possible to configure a plurality of logical units of different RAID levels in a plurality of drives constituting a logical group.

However, in one logical group, for example, a parity group consisting of 5 data and 1 parity, a parity group consisting of 3 data and 1 parity, 5 data and 2 parity. There is a problem that parity groups of various configurations such as parity groups cannot be mixed. In addition, since failure recovery is performed in units of logical groups, failure in accordance with the characteristics of each RAID level (probability of data loss, performance at the time of failure) for a plurality of logical units of various RAID levels set in the logical group. There is no indication that recovery processing should be performed.

An object of the present invention is to improve a disk array. Another object of the present invention is to provide a disk array device and a disk array partitioning method in which parity groups of various configurations can be mixed in one logical group. Further, the object of the present invention is to
To provide a disk array device and a disk array partitioning method that perform failure recovery processing in a distributed manner according to characteristics that differ depending on the AID level and the configuration of a parity group, such as the probability of data loss and the degree of performance degradation at the time of failure. is there.

[0028]

According to the present invention, there is provided a disk array device including a disk device including a plurality of drives forming a logical group, and a controller for managing the disk device. An area for storing a first parity group composed of (≧ 1 integer) data and j (j is an integer of j ≧ 1) error correction data created from the data; An area for storing a second parity group having a different configuration from one parity group is mixed in the logical group. The parity group also includes a redundant parity group (i = j = 1).

Further, according to the present invention, in a disk array device including a disk device including n drives forming a logical group and a control device for managing the disk device, each drive in the logical group is partitioned. Partition, and select any number of arbitrary partitions of n or less m drives, set a partition group by the selected partition, and the logical group is configured by a plurality of partition groups in which m is different from each other. It is characterized by being done.

Furthermore, in response to an input / output request of data from a higher-level device, a disk array device comprising a disk device which stores or stores the data and a control device which manages the disk device. , The disk device is composed of a large number of drives, these drives are divided into logical groups of two or more n drives, and each drive in each logical group is partitioned by a partition and m or less It is characterized in that an arbitrary number of arbitrary partitions of arbitrary drives are selected, a partition group is set by the selected partitions, and the logical group is composed of a plurality of partition groups having mutually different m.

[0031]

As described above, according to the present invention, the parity group level is made variable among the parity groups constituting the logical group, and the logical group is constituted by several parity group levels. That is, within a logical group,
Multiple types of parity groups can be mixed. Each parity group is characterized by reliability and performance. Therefore, depending on the configuration of the parity group, 2
The probability of data loss due to the failure of the second drive is different, and it is classified as the parity group level. This can be said to be the same when the performance is deteriorated at the time of failure.

Therefore, when a failure occurs in the first drive, for example, recovery processing is sequentially performed from a parity group having a low parity group level (high probability of data loss when a second failure occurs). Do it.

In this way, the parity group level is made variable among the respective parity groups constituting the logical volume, the logical groups are constituted by several parity group levels, and the recovery processing is performed sequentially in the ascending order of the parity group levels. It is possible to secure reliability and perform recovery processing efficiently, and it is also possible to suppress performance degradation at the time of failure.

[0034]

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

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

The disk array of this embodiment is roughly composed of a disk array controller (hereinafter referred to as ADC) 2 for controlling RAID level 5 and a logical group 10. The CPU 1 is a host device that issues a read command and a write command to the disk array.

The logical group 10 includes m drives 12
And drive unit paths 11-1 and 11-2 connecting each drive 12 and the ADC 2. The number of the drives 12 is not particularly limited in order to obtain the effects of the present invention.

Within the logical group 10, the set of data for which parity is created and the parity created from these data are combined to form a parity group. Conventionally, this logical group 10 has been the recovery unit for drive failure. In this embodiment, parity groups having different configurations are mixed in this logical group 10 and failure recovery is performed according to the characteristics of each parity group. This will be described in detail later.

Next, the internal structure of the ADC 2 will be described with reference to FIG.

The ADC 2 roughly comprises a channel path director 3, channel paths 8-1 and 8-2, a cache memory 16 and drive paths 9-1 and 9-2. Also, the path is roughly two clusters 7-1 and 7-
It is divided into two.

The channel paths 8-1 and 8-2 are paths between the channel path director 3 and the cache memory 16. The cache memory 16 is a semiconductor memory that is non-volatile by battery backup or the like. The drive paths 9-1 and 9-2 are paths between the cache memory 16 and the drive 12. The respective channel paths 8-1, 8-2 and the drive paths 9-1, 9-2 are connected via a cache memory 16.

The cache memory 16 stores data and an address table, and also serves as a work area when creating parity. The cache memory 16 and the address table therein are shared by all the clusters 7 in the ADC 2. Details of the address table will be described later with reference to FIG.

This address table is cached from a certain one or more drives 12 in the logical group 10 by the microprocessor (MP) 20 in the channel path 8 when the system is powered on. It is automatically read into the memory 16 without any knowledge of the CPU 1. On the other hand, when the power is turned off, the CPU 20 sets the address table in the cache memory 16 to a predetermined location in the drive 12 that has been read by the MP 20.
Automatically store without knowledge of.

The command issued by the CPU 1 is input to the channel path director 3 of the ADC 2 through the external interface path 4. When a command is issued from the CPU 1, the channel path director 3 in the ADC 2 first determines whether or not the command can be accepted.

Specifically, as shown in FIG. 1, the CPU 1
The command sent from the ADC 2 to the ADC 2 is first captured by the interface adapter (hereinafter, referred to as IF Adp) 5. Next, the MP 20 checks whether or not there is a usable path in the external interface path 4. If there is a usable external interface path 4, the MP 20 switches the channel path switch 6 to perform command acceptance processing. If the command cannot be accepted, a response indicating that the command cannot be accepted is sent to the CPU 1.

After the command can be accepted in this way, a read process or a write process (new or update) as described later is started. Before describing a specific read or write process, first, a parity group level setting method and an address conversion method in the initial setting will be described.

(Parity group level setting method)

FIG. 2 shows the logical group 1 in this embodiment.
It is a data allocation diagram in 0. In this embodiment, as shown in this figure, six drives 12 of SD # 1 to SD # 6 are used.
The logical group 10 is constructed by

In FIG. 2, in the logical group 10,
A parity group that forms level 5 with 5 pieces of data and 1 parity (5D + 1P), a parity group that forms level 5 with 3 pieces of data and 1 parity (3D + 1P), and a parity group by duplication And three types of parity groups are mixed. For these 3 types of parity groups, 5D + 1P, 3D
The reliability is improved in the order of + 1P and duplexing, and the performance degradation at the time of failure becomes smaller in this order. Therefore, the parity groups are classified as having a higher parity group level in order of reliability and performance.

In order to obtain a desired usage environment, the user
At the stage of initial setting, how many levels of parity groups are prepared in the logical group 10 is preset. This setting is instructed to the ADC 2 by the user.
In the ADC 2, the MP 20 receives this instruction. In response to the instruction, the MP 20 creates an address table in the cache memory 16.

FIG. 4 shows the structure of the address table created in the cache memory 16. Here, an address table in the case where three types of parity groups of 5D + 1P, 3D + 1P, and duplex are provided as shown in FIG. 2 is shown. The address table is shown in (a), (b) and (c) of FIG.
As shown in, it is created corresponding to the parity group level.

The address table corresponding to each parity group level 21 has a data name 22, a cache address 23, and a data drive number (DDrive No.).
24, failure flag 25, parity drive number (PDr
iv No. ) 26, and Addr27 in SCSI
It is composed of each field of.

In the data name 22, a logical address designated by the CPU 1 is registered. A portion where the data name 22 is not registered indicates a free area. In the cache address 23, when the data corresponding to the data name 22 exists in the cache memory 16, the address in the cache memory 16 is registered. On the contrary, if the cache address 23 is not registered (blank), it means that the data of the data name 22 does not exist in the cache memory 16.

Data drive number (DDrive N
o. ) 24, the number of the drive 12 in which the data of this data name 22 is stored is registered. The failure flag 25 is turned on (1) when the drive 12 storing this data has a failure. If the drive is normal, the failure flag 25 is off (0). Parity drive number (PDrive No.) 2
In 6 is registered the number of the drive 12 storing the parity in which the data of the data name 22 is involved in the parity group.

Data name 2 is stored in Addr 27 in SCSI.
The physical address in the drive 12 in which the data of 2 and the parity related to this are stored is registered. The physical address indicated by the Addr 27 in the SCSI determines the position of the cylinder to which the track storing the data belongs and the track storing the data in the cylinder, as shown in FIG. It is represented by the head address and the position of the record within that track.

Specifically, the number of the drive 12 in which the requested data is stored is the data drive number (DDr
iv No. ) 24, the cylinder address that is the cylinder number in the drive 12, the head address that is the number of the head that selects the track in the cylinder, and the record address that indicates the position of the record in the track. It is registered in Addr 27 in SCSI.

In the present embodiment, in each drive 12 constituting the logical group 10, the data constituting the parity group and the parity created from these data are stored at the same SCSI Addr 27 position.

Specifically, for example, in FIG.
For the data whose data name 22 is Data # 1 to Data # 5, Addr27 in SCSI is DADR1. This is Data stored in the drive SD # 1.
a # 1 and Data stored in the drive SD # 2
# 2 and Data # stored in the drive SD # 3
3 and Data # 4 stored in the drive SD # 4
And Data # 5 stored in the drive SD # 5
And the parity stored in the drive SD # 6,
It is shown that the data is stored at the address DADR1 of each drive. A set of these data forms a parity group.

In this embodiment, the data and parity forming the parity group are assigned to the same SCSI Ad.
It was stored in dr27, but Addr in SCSI
27 may be different for each drive. In that case, in addition to the drive numbers, the SCSI Addr in each drive may be stored in the drive numbers 24 and 26.

In this embodiment, the parity group is set in cylinder units. As shown in FIG. 17, drive 1
The second disc has concentric tracks. Data is recorded on this track. The read / write heads corresponding to each disk are all attached to one actuator, and all the read / write heads move uniformly with the movement of this actuator. A set of tracks on each disk positioned by one movement of the actuator is called a cylinder.

When a parity group is set for each parity group level, it is set for each set of cylinders as shown in FIG. In FIG. 18, a parity group having a parity group level of 5D + 1P is set in the set of cylinders on the outer peripheral side (hatched portion), and a set of parity group levels of 3D + 1P and duplexing parity is set in the set of cylinders on the inner peripheral side (the blackened part). The groups are set mixedly.

At the time of initial setting, the user selects the DDrive No. corresponding to the area (set of cylinders) assigned to each parity group level. 24 and P Drive N
o. 26 and their Addr 27 in SCSI are shown in FIG.
Reserved on the address table of.

In this embodiment, the parity group is set in the cylinder unit as described above, but it is clear that the effect of the present invention can be obtained even if the parity group is set in the track unit and the record unit. There is no unit restriction.

In this embodiment, the parity group level is set to 5D + 1P, 3D + 1P, and duplex, but this setting can be freely set depending on the environment of use. For example, if a more reliable parity group is required, a parity group level of 4D + 2P can be set, or a parity group level such as 2D + 1P can be set. When two parities are provided as in 4D + 2P, two fields of the parity drive number (PDrive No.) 26 in FIG. 4 may be provided.

Further, the ratio of the parity group level in the parity groups forming the logical group 10 can be freely set depending on the usage environment. For example, in FIG. 2, it is possible to reduce the number of parity groups (allocation amount) at the parity group level of 5D + 1P and increase the allocation amount of 3D + 1P and redundant parity groups, or to set a parity group of 4D + 2P instead. Is.

As described above, according to the present invention, in consideration of the environment in which the disk array is used at the time of initial setting, the type of the parity group level of the parity group forming the logical group 10 of the disk array and the configuration contents Moreover, it is possible to set the ratio freely.

(Address conversion method)

Next, an address conversion method using the address table of FIG. 4 will be described.

The CPU 1 of FIG. 1 designates the data name 22 as a logical address and issues a write command or a read command. In response to this, the MP 20 of the ADC 2 refers to the address table of FIG. 4 with the designated logical address to determine the drive number 24, the SCSI Addr 27, and the cache address 23 in which the data is actually stored. To do.

For example, when a request for Data # 2 is issued from the CPU 1, the location of the data is found to be the SCSI of the drive SD # 2 from the address table of FIG.
It can be seen that the Addr 27 is the position of DADR1. In this way, the data name as the logical address is converted into the physical address.

At this time, since CADR1 is registered in the cache address 23 of Data # 2 in the address table, this data is stored in the cache memory 1.
It can be seen that it exists in CADR1 within 6. If the cache address 23 is not registered, the data does not exist in the cache memory 16. Further, regarding the parity involving this Data # 2, from the address table of FIG. 4, the Addr 27 in SCSI of the drive 12 whose parity drive number (PDrive No) 26 is SD # 6 is stored at the position of DADR1. You can see that

In this way, the MP 20 converts the logical address 22 designated by the CPU 1 into the physical address of the drive 12 which actually reads / writes, and then issues a read or write request to the physical address. To do.

Next, a processing method for performing such address conversion and specifically reading or writing data will be described.

(New writing process)

First, a method of newly writing data will be described with reference to FIG. Further, FIG. 6 shows a processing flow of the MP20 when newly writing data.

First, description will be made with reference to FIG. Upon receiving the command from the CPU 1, the MP 20 of the ADC 2 checks whether the command can be processed, and if possible, returns a response indicating that the command can be processed to the CPU 1. The command issued by the CPU 1 is taken into the ADC 2 via the IF Adp 5 and is decoded by the MP 20 as a read request or a write request. In the case of a write request, it is processed as follows.

After receiving the response that the processing is possible, the CPU 1 transfers the write data to the ADC 2. At this time, in the ADC 2, the channel path switch 6 of the channel path director 3 connects the external interface path 4 and the IF Adp 5 to the channel path 8 according to the instruction from the MP 20, and establishes the connection between the CPU 1 and the ADC 2. After establishing the connection between the CPU 1 and the ADC 2, the data transfer from the CPU 1 is accepted.

A logical address (data name) and write data are transferred from the CPU 1. The channel interface circuit (CH IF) 13 performs protocol conversion on these data transferred from the CPU 1 according to an instruction from the MP 20. As a result, the data from the CPU 1 is
The speed is adjusted to the processing speed in DC2. In particular,
When the channel interface between the CPU1 and the ADC2 is an optical interface, CH IF13
Converts the protocol of the optical interface into a protocol for electrical processing in the ADC 2.

After completion of the protocol conversion and speed control in the CH IF 13, the data is transferred to the data control circuit (D
CC) 14, the data is transferred to the cache adapter circuit (C Adp) 15, and the data is transferred to C Adp.
It is stored in the cache memory 16 by 15.

The C Adp 15 is a circuit for reading and writing data from and to the cache memory 16 in accordance with instructions from the MP 20.
It is a circuit that monitors the state of, and performs exclusive control for each read and write request.

Referring to the flowchart of FIG. 6, MP2
The processing procedure by 0 will be described in more detail. MP20
Recognizes the command of the write request, and recognizes that the data to be written is new data to be written for the first time (step 30), the MP 20 uses CP as a logical address.
The process of registering the data name sent from U1 in the address table is started.

First, the MP 20 recognizes the parity group level (step 31). That is, when the parity group level is designated from the upper level, the address table to be used is determined so as to be stored in the area of the parity group level. Here, it is assumed that when a write request command is given, the parity group level for writing data is designated by the CPU 1.

Specifically, when the CPU 1 instructs to store in the area of the parity group level of 5D + 1P, the address table of FIG.
In P, the address table of FIG. 4B is used, and in the case of duplication, the address table of FIG. 4C is used.

After determining the address table in this way, the MP 20 searches for an empty area in which the data name is not registered in the address table (step 3).
2). In this embodiment, when the data name 22 is not registered in the address table of FIG. 4, the DDrive No. The area of the Addr 27 in SCSI in the drive 12 of 24 is a free area in which data is not stored.

There are cases where data is not written in this empty area from the initial setting stage, and cases where data has been written before but this data is no longer needed and is deleted. Data is deleted by the MP 20 by deleting the data name 22 in the address table of FIG. In addition, as for the parity at this time, all the bits in the area where the data is not stored are 0.
No. of the PDrive No. specified in the address table. Drive 12 of 26
Is stored in the same position of Addr 27 in SCSI as the data.

Next, the new write data is stored in the cache memory 16 (step 33). The data flow up to storage in the cache memory 16 has been described above with reference to FIG. Then, the data name 22 is registered in the empty area of the address table found in step 32, and the address in the cache memory 16 storing the new write data is registered in the cache address 23 to register the new write data. Is performed (step 34).

Next, the MP 20 updates the parity by using the new write data, and starts the process of storing the new write data and the new parity in the drive 12 in the logical group 10. Hereinafter, the parity before update will be referred to as the old parity, and the updated parity will be referred to as the new parity.

First, the MP20 sends the DDrive No. corresponding to the new data registered in the address table.
29 and PDrive No. 34 and SCSI Addr2 which is a physical address in those drives.
7 is recognized (step 35). And DDriv
e The drive of No. 29 is instructed to write the new data stored in the cache memory 16 (step 36).

Further, the drive of PDrive No. 34 is instructed to update the parity. That is, P
The old parity is read from the drive of Drive No. 34 (step 37), the new parity is created from the old parity and the new write data (step 38), and the new parity is written to the drive of P Drive No. 34 (step 39). .

Using the example of FIGS. 2 and 4, step 3
The process after 5 will be specifically described.

In FIGS. 2 and 4, MP20 is D
Drive No. 24 is S of the drive 12 of SD # 2
Add D27 in CSI to the position of DADR1 and newly add D
It is assumed that when the data # 2 is written, it is recognized. Also, MP
20 is the storage position (PDrive of the parity corresponding to this new write data from the address table of FIG.
No. 26 is A in SCSI of drive 12 of SD # 6
ddr27 recognizes the position of DADR1).

After recognizing the data to be newly written and the physical address in the logical group 10 of the parity involved after writing this data, the MP 20 performs the writing process of the data to be newly written to the drive SD # 2. Then, the parity update process is started for the drive SD # 6.

Parity update processing will be described with reference to FIG.

In the parity update process, first, the MP20
But drive interface (Drive IF) 1
8 is instructed to the drive 12 to issue an old parity read request. The Drive IF 18 issues a read command via the drive unit path 11 in accordance with the SCSI read processing procedure.

In the drive 12 to which the read command is issued from the Drive IF 18, the seek and rotation waiting access processing is performed to the designated SCSI Addr 27. After the access processing in the drive 12 is completed, the drive 12 reads the old parity and drives the drive unit path 11 to drive the drive parity.
ve IF18.

The Drive IF 18 transfers the transferred old parity to the cache adapter circuit (C Adp) 17 on the drive 12 side. C Adp17
Stores the old parity in the cache memory 16. At this time, the C Adp 17 reports to the MP 20 that the old parity is stored in the cache memory 16.

In this way, after reading the old parity into the cache memory 16, the MP 20 instructs the parity generation circuit (PG) 19 to create a new parity. The PG 19 calculates the exclusive OR of the new write data stored in the cache memory 16 and the old parity, and creates the updated new parity. The created new parity is stored in the cache memory 16.

After the storage of the new parity in the cache memory 16 is completed, the MP 20 controls to write the new write data (Data # 2 in this example) to the drive SD # 2 and the new parity to the drive SD # 6. To do.
The storage location in each drive is A in SCSI.
It is DADR1 shown in ddr27.

Next, the process of writing new write data and new parity to the drive 12 will be described.

First, MP20 is driven by Drive IF18.
The drive 12 is instructed to issue a write request (write of new write data or new parity). The Drive IF 18 issues a write command to the drive 12 via the drive unit path 11 according to the SCSI write processing procedure.

In the drive 12 to which the write command is issued from the Drive IF 18, the access process for seeking and rotating to the instructed Addr 27 in SCSI is performed. After the access process in the drive 12 is completed, the cache adapter circuit (C Adp)
17 reads new write data or new parity from the cache memory 16 and transfers it to the Drive IF 18.

The Drive IF 18 transfers the transferred new write data or new parity to the drive 12 via the drive unit path 11, and writes the new write data or new parity to the address of the drive 12. At this time, C Adp17
Informs the MP 20 that new write data or new parity has been stored in the drive 12.

(Write processing)

Next, the case of updating in which the data already written in the drive 12 is rewritten to new data will be described. FIG. 7 is a flowchart showing the procedure of the update writing process.

A data update request is generated (step 4)
0), the MP 20 refers to the address table from the logical address (data name) designated by the CPU 1, and DDri of the drive 12 in which data and parity are stored.
ve No. 24 and PDrive No. 26, the SCSI Addr 27 which is a physical address in the drive 12, the cache address 23, and the failure flag 25 are recognized (step 41).

For example, in the example of FIGS. 2 and 4, CPU1
From drive SD # 2 SCSI addr 27 is DA
It is assumed that a write request for updating is issued to Data # 2 of DR1. At this time, the MP 20 first adds the DD of the data (old data) updated by the address table.
live No. 24, Addr 27 in SCSI, cache address 23, and PDrive No. of the parity to be updated (old parity). Recognize 26.

Next, the failure flag 2 in the address table for the drive 12 in which the data is stored.
It is determined whether or not 5 is off (0). If the failure flag 25 is on (1), the process proceeds to the update processing at the time of failure (FIG. 8). If the failure flag 25 is off (0), the MP 20 recognizes that the drive 12 is normal and processes as follows.

The new data to be updated is stored in the cache memory 16 from the CPU 1 in the same manner as the new write data at the time of writing the new data (step 43).
Then, the MP 20 registers the address in the cache memory 16 storing the new data to be written, in the cache address 23 of the address table (step 4).
4).

If the old data is not in the cache memory 16, the old data and the old parity are read out and stored in the cache memory 16 (steps 48 and 45). If the old data is in the cache memory 16, step 48 is skipped and only the old parity is read from the drive 12 and stored in the cache memory 16 (step 45). The method of reading the old data and the old parity from the drive 12 and storing them in the cache memory 16 at this time is the same as the method of reading the old parity from the drive 12 to the cache memory 16 in the parity update processing at the time of new writing described above. is there.

DDrive No. for storing data. Two
In the drive of No. 4, after reading the old data, the new data is written to the position of Addr 27 in SCSI of the drive (step 49). Regarding parity, the old data read in this way, the old parity and the new data to be written are used to calculate the exclusive OR in the same way as the parity update processing at the time of new writing described above, and the new parity after updating is calculated. It is created and stored in the cache memory 16 (step 46). And, the new parity created is PDri
ve No. Addr27 in SCSI of 26 drives
Is written in the position (step 47).

Writing new data and new parity
It is performed in the same manner as the above-described new writing.

(Processing for writing to the failed drive 12)

Next, when updating the data already written in the drive 12 with new data,
A processing method when a failure occurs in the drive 12 storing the old data to be updated will be described. FIG. 8 is a flowchart showing the procedure of the update process in the event of such a failure.

Before the processing of FIG. 8 is started, MP20
7 refers to the address table from the logical address (data name) specified by the CPU 1 in step 41 of FIG. 7, and the drive 1 in which data and parity are stored
No. 2 DD Drive No. 2 24 and PDrive No.
26, SCSI addr 27 which is a physical address in the drive 12, cache address 23,
And the failure flag 25 is recognized. Step 42
Then, the failure flag 25 in the address table for the drive 12 in which the data is stored is turned on (1).
Then, the MP20 recognizes that this drive 12 is abnormal,
The process of FIG. 8 is started.

For example, in the example of FIGS. 2 and 4, CPU1
From the drive SD # 1 SCSI addr 27 is DA
It is assumed that a write request for updating is issued to Data # 1 of DR1. At this time, the MP 20 first adds the DD of the data (old data) updated by the address table.
live No. 24, Addr 27 in SCSI, cache address 23, and PDrive No. of the parity to be updated (old parity). Recognize 26. At this time, in the address table, the old data Data #
Fault flag 25 of drive SD # 1 in which 1 is stored
Is on (1), the MP 20 recognizes that a failure has occurred in the drive SD # 1, and starts the write processing at the time of failure shown in the flowchart of FIG.

In FIG. 8, first, the new data to be updated is
The data is stored in the cache memory 16 from the CPU 1 in the same manner as the new write data at the time of writing new data (step 50). Next, the MP 20 registers the address storing the new data to be written in the cache memory 16 in the cache address 23 of the address table (step 51).

Next, the MP 20 recognizes the addresses of all the data and the parity related to the recovery process of the old data in the parity group by the address table (step 52). And the MP20 is the drive 1
2, these data and parity are read,
It is stored in the cache memory 16 (step 53). At this time, among these data and parity, the one whose address is registered in the cache address 23 in the address table is already in the cache memory 16, and the MP 20 performs the reading process from the drive 12. Absent.

The method of storing these data and parity from the drive 12 in the read cache memory 16 is the same as the write processing at the normal time, and the drive 12 of the old parity in the parity update processing at the time of new writing described above. From the cache memory 16 to the cache memory 16.

In this way, the exclusive OR calculation is performed on all the data read and the old parity and the new data to be written, as in the normal write process, as in the parity update process at the new write described above. The updated new parity is created and stored in the cache memory 16 (step 54). When the parity update is completed in this way, only the new parity is updated in the drive 12 by the same method as the parity update process at the time of new writing described above.
SCS of (PDrive No. 26 drive)
By storing it in the position of Addr 27 in I, only the parity is updated (step 55).

(Reading process)

Next, the case of reading the data already written in the drive 12 will be described. Figure 9
FIG. 6 is a flowchart showing the procedure of such data read processing.

When a data read request is generated (step 56), the MP 20 refers to the address table from the logical address (data name) designated by the CPU 1 and drives the drive 12 in which the data to be read is stored.
ve No. 24, SCSI A in the drive 12
The ddr 27 and the failure flag 25 are recognized (step 57). Then, the failure flag 2 in the address table for the drive 12 in which the data is stored
It is determined whether or not 5 is off (0) (step 58). If the failure flag 25 is on (1), the process proceeds to the read processing at the time of failure (FIG. 10). If the failure flag 25 is off (0),
The MP 20 recognizes that the drive 12 is normal and processes as follows.

First, the MP 20 checks the cache address 23 for the data name in the address table (step 59) and determines whether or not there is data to be read in the cache memory 16 (step 60).
If the address is registered in the field of the cache address 23 and the data to be read is stored in the cache memory 16 (cache hit), M
The P20 starts control of reading the data from the cache memory 16 (step 61). If it does not exist in the cache memory 16 (cache miss), the drive 12 starts control to read the relevant data therein (from step 62).

At the time of a cache hit, the MP 20 converts the logical address (data name) designated by the CPU 1 in the address table into the cache address 23 in the cache memory 16 in which the data is stored,
The data is read out to the cache memory 16 (step 61). Specifically, the data is read from the cache memory 16 by the cache adapter circuit (C Adp) 15 under the instruction of the MP 20.

The data read by the C Adp 15 is transferred to the channel interface circuit (CH IF) 13 under the control of the data control circuit (DCC) 14. The CH IF 13 performs a process of converting the read data into the protocol of the channel interface in the CPU 1, and adjusts the speed to a speed corresponding to the channel interface. After the protocol conversion and speed adjustment in the CH IF 13, in the channel path director 3, the channel path switch 6 selects the external interface path 4, and the IF Adp 5 transfers the data to the CPU 1.

On the other hand, if it is determined in step 60 that there is a cache miss, the MP20 uses the address table to specify the DDrive N corresponding to the logical address (data name) designated by the CPU 1 as in the cache hit.
o. 24, S which is the physical address in the drive
The CSI Addr 27 and the failure flag 25 are recognized (step 62).

Next, the MP 20 instructs the Drive IF 18 to issue a read request to the drive 12 for the physical address (step 6).
3). The Drive IF 18 issues a read command via the drive unit path 11 in accordance with the SCSI read processing procedure.

In the drive 12 to which the read command has been issued from the Drive IF 18, the access process for seeking and rotating to the instructed Addr 27 in SCSI is performed. After the access process in the drive 12 is completed, the drive 12 reads the data and drives through the drive unit path 11.
Transfer to IF18.

In Drive IF 18, the transferred data is transferred to the cache adapter circuit (C Adp) 17 on the drive 12 side, and in C Adp 17,
Data is stored in the cache memory 16 (step 6)
3). At this time, C Adp17 is
MP2 indicates that the data is stored in the cache memory 16.
Report to 0. Based on this report, the MP 20 registers the address in the cache memory 16 storing the data in the cache address 23 of the logical address (data name) to which the CPU 1 of the address table has issued the read request (step 65). After that, the data is transferred to the CPU 1 in the same procedure as when the cache hits (step 64).

(Reading process from faulty drive 12)

Next, the case of reading the data written in the failed drive 12 will be described. Figure 10
6 is a flowchart showing a procedure of a read process from a faulty drive.

Before the processing of FIG. 10 is started, MP20
Is the drive 12 in which the data to be read is already stored by referring to the address table from the logical address (data name) designated by the CPU 1 in step 57 of FIG.
Drive No. 24, SC in the drive 12
The SI Addr 27 and the failure flag 25 are recognized. In step 58, the failure flag 25 in the address table for the drive 12 in which the data is stored is determined. Fault flag 25 is on (1)
Then, the MP20 recognizes that this drive 12 is abnormal,
The process of FIG. 10 is started.

First, the MP 20 checks the cache address 23 for the data name in the address table (step 66), and determines whether or not there is data to be read in the cache memory 16 (step 67).
If the address is registered in the field of the cache address 23 and the data to be read is stored in the cache memory 16 (cache hit), M
The P20 reads the data from the cache memory 16 and transfers the data to the CPU 1 (step 6).
8). If it does not exist in the cache memory 16 (cache miss), control for reading the data in the drive 12 is started (step 69 and after).

In the case of a cache miss, the MP 20 uses the address table to determine the DDrive No. of the drive 12 in which all the data and the parity involved in the recovery process of the old data in the parity group are stored. 24 and PDrive No. 24. 26, the SCSI addr 27 which is a physical address in the drive, and the failure flag 25 are recognized (step 69).

Next, the MP20 is configured to drive each drive 1
2, these data and parity are read,
It is stored in the cache memory 16 (step 70). At this time, among these data and parity, it is assumed that the address whose address is registered in the cache address 23 in the address table already exists in the cache memory 16, and the MP 20 determines that the drive 12
The read processing from is not performed.

The method of reading these data and parity from the drive 12 and storing them in the cache memory 16 is the same as the method of reading the old parity from the drive 12 to the cache memory 16 in the parity update processing at the time of new writing described above. Is the same.

Next, the MP 20 instructs the PG 19 to restore the data. The PG 19 performs an exclusive OR calculation using the data and parity read in step 70, restores the data stored in the failed drive 12, and stores it in the cache memory 16 (step 71). At this time, the C Adp 17 reports to the MP 20 that the data is stored in the cache memory 16. The MP 20 registers the address in the cache memory 16 storing the data in the cache address 23 of the logical address (data name) to which the CPU 1 of the address table has issued the read request (step 72).
After that, the CPU 1 performs the same procedure as the cache hit.
The data is transferred to (step 71).

(Failure recovery processing)

Next, a procedure for recovering the data in the failed drive 12 when the failure occurs in the drive 12 will be described. FIG. 11 is a flowchart showing a procedure for such failure recovery.

First, when the MP 20 recognizes that a failure has occurred in any of the drives 12 in the logical group 10 (step 73), all the items on the address table relating to the failed drive 12 are failed. The flag 25 is turned on (1). For example, when a failure occurs in the drive SD # 1, the MP20 turns on (1) all the failure flags 25 in the section related to the drive SD # 1, as shown in the address table of FIG. Then, the process proceeds to step 75 and the subsequent steps.

Before the description of the processing after step 75, the failure recovery method of the present embodiment will be described using an example.

In the examples shown in FIGS. 2 and 4, five data and one parity (5D + 1P), three data and one parity (3D + 1P) are included in the drive 12 that constitutes the logical group 10. , And dual parity group level parity groups exist. D
ADR1, 2, and 3 have a parity group level of 5D + 1
In the P parity group, DADRs 4, 5 and 6 have a parity group level of 3D + 1P and a dual parity group.

At these three types of parity group levels, the reliability is high in the order of 5D + 1P, 3D + 1P, and duplication, and the performance at the time of failure is also high.

Specifically, 5D + of DADR1, 2, 3
In the 1P parity group, if the first drive 12 fails and any drive 12 in the logical group 10 fails before the recovery process, data will be lost. On the other hand, in the 3D + 1P parity group of DADR 4, 5, and 6, even if a failure occurs in the first drive 12 in the logical group 10 and another failure occurs in the other drive 12 before the recovery processing, the parity The data is not lost unless the drive 12 stores the data forming the group. Further, regarding the redundant parity group, the drive 12 of the redundant pair is used.
Otherwise, no data will be lost.

As described above, the probability of data loss due to a failure in the second drive 12 differs depending on the configuration of the parity group. Conversely, parity groups having different probabilities are classified as different parity group levels.

Further, the classification by the parity group level can be applied to the performance deterioration at the time of failure, as shown below. For example, as described in the writing / reading process at the time of failure, when a writing / reading request is issued to the drive 12 in which the failure has occurred, the reading request must be issued to the five drives 12 in 5D + 1P. I won't. However, 3D +
In 1P, only one read request is required for the three drives 12, and only one is required for duplication. As described above, when a read request or a write request is issued to the failed drive 12, the number of drives 12 that must issue the read request in the recovery process is five.
Since D + 1P, 3D + 1P, and duplex are less in order, the performance at the time of failure is higher in this order.

Therefore, in this embodiment, a failure occurs in the first drive 12 and the drive 12 in which this failure has occurred.
11 and FIG. 12 when performing a recovery process for restoring the data in the internal parity group, the parity group having a lower parity group level (having a higher probability of data loss (data loss probability) when a second failure occurs) is used. As shown in, the recovery process is sequentially performed.

At present, as the capacity of the drive 12 increases, the file server, which is one of the promising application destinations of the disk array, has become larger in size, and has been used by many users. For this reason, when a failure occurs in the disk of the file server and the operation is stopped for a long time to recover from the failure, the work of many users is stopped, which causes a great damage. For this reason, it is desirable to stop and perform recovery processing when the number of users who use the file system is as small as possible rather than during the time when many users are using it.

In the present invention, high reliability is ensured and performance deterioration at the time of failure is suppressed to a minimum, and recovery processing is performed not in a concentrated manner but in a dispersed manner. Therefore, a specific recovery method is shown below.

In FIGS. 2 and 4, drive SD #
Suppose that 1 fails. For drive SD # 1, S
Addr27 in CSI is Data # at the position of DADR1.
1 for DADR2, Data # 6 for DADR3, and Da for DADR3
ta # 11, DADR4, Data # 16, DAD
Data # 20 is stored in R5, and parity is stored in DADR6. Data # 1, Data # 6
Data # 11 belongs to the parity group of the 5D + 1P parity group level, Data # 16 belongs to the parity group of the redundant parity group level, and Data # 20 and parity belong to the parity group of the 3D + 1P parity group level. There is.

Therefore, the MP 20 first considers the recovery of data belonging to the parity group of the parity group level of 5D + 1P, which has a high probability of data loss and low performance in the event of a failure. Specifically, the recovery means that the drive SD # 1 in which the failure has occurred is replaced with a normal drive 12 (for example, a message prompting the replacement of the drive is output and the user is allowed to replace the drive). In the case where a spare drive 12 is prepared in advance for a failure, the processing is switched to this spare drive 12, for example.

Returning to FIG. 11, the procedure of the recovery process as described above will be described.

First, the MP20 is
The number of read and write requests (number of IOs) issued to the logical group 10 to which the failed drive 12 which is going to perform recovery processing belongs is checked (step 75). Then, it is determined whether or not the number of IOs is less than or equal to a preset value (step 76).

When the number of IOs exceeds the set value and a read or write request is issued to the drive 12 in which a failure has occurred, as described in the above-mentioned read processing or write processing to the failed drive, A read or write request from the higher order is answered using the parity (step 77). On the other hand, if the number of IOs is less than or equal to the set value, the process of recovering the data belonging to the parity group of the parity group level of 5D + 1P is performed and the failure flag is turned off (step 78). The recovery process restores individual data as it did in the reading to the failed drive 12.

The restoration of individual data will be specifically described. First, the MP 20 restores Data # 1. MP20 uses the address table to
2, 3, 4, 5 and the DDrive No. in which the parity created from these data is stored. 24 and PDr
iv No. 26, the SCSI addr 27 which is a physical address in the drive 12, and the failure flag 25.

Next, the MP20 operates on each drive 1
2, these data and parity are read,
It is stored in the cache memory 16. At this time, among these data and parity, in the address table of FIG. 4, addresses are already registered in the cache address 23 for Data # 2, 4, and 5, so these data have already been stored in the cache memory 16. Existing. In that case, the MP 20 does not read from the drive 12.

The method of reading these data and parity from the drive 12 and storing them in the cache memory 16 is the same as the method of reading from the drive 12 to the cache memory 16 described above.

The MP 20 instructs the PG 19 to restore the data and store it in the cache memory 16. PG1
9 performs an exclusive OR calculation with the data and the parity read as described above, restores Data # 1 stored in the faulty drive 12, and stores it in the cache memory 16. At this time, C Adp17 becomes MP2
0 is reported to store data in the cache memory 16.

Based on this report, the MP 20 registers the data storage address in the cache memory 16 in the cache address 23 of Data # 1 in the address table,
The failure flag 25 is turned off (0). The data thus restored is stored in the replaced normal drive 12, for example. Data # 6 and 11 are similarly restored and stored in the replaced normal drive 12.

As described above, after the data belonging to the parity group at the parity group level of 5D + 1P is restored and recovered, the probability of data loss is reduced and the performance at the time of failure is improved.

When the recovery of the parity group level of 5D + 1P is completed, the MP20 again causes the faulty drive 12 (actually, the replacement to the normal drive or the replacement to the spare drive has already been performed in step 78) from the upper CPU 1. However, at this point, only 5 are restored.
Since only D + 1P and other parity group levels have not been restored, the number of read and write requests (the number of IOs) issued to the logical group 10 to which the other drive belongs is referred to for the sake of convenience. IO
It is determined whether the number is less than or equal to a preset value (step 79).

When the number of IOs exceeds the set value and a read or write request is made to the 3D + 1P of the faulty drive 12 or the redundant parity group, the above-mentioned read processing or write processing to the faulty drive will be described. As described above, the parity is used to answer the read or write request from the higher order (step 8).
0). On the other hand, if the number of IO is less than the set value, 3D + 1
The data belonging to the parity group of the P parity group level is recovered and the failure flag is turned off (step 81). The recovery processing method is similar to the parity group level of 5D + 1P.

As described above, after the data belonging to the parity group of the parity group level of 3D + 1P is restored and recovered, the probability of data loss further decreases and the performance at the time of failure further improves.

When the recovery of the parity group level of 3D + 1P is completed, the MP 20 checks the number of read and write requests (the number of IOs) issued from the upper CPU 1 to the logical group 10 to which the failed drive 12 belongs, and the IO It is determined whether the number is less than or equal to a preset value (step 82).

When the number of IOs exceeds the set value and a read or write request is made to the redundant parity group of the faulty drive 12, as described in the above-mentioned read or write process to the faulty drive. Then, the read or write request from the upper layer is answered using the drive which is not the redundant drive (step 83). On the other hand, when the number of IOs is equal to or less than the set value, the process of recovering the data belonging to the parity group of the redundant parity group level is performed and the failure flag is turned off (step 84).

The recovery processing method is, for example, drive SD #
Data # 16 is read from the drive SD # 2 in which the duplicated data of Data # 16 stored in No. 1 is stored in the cache memory 16 to
From the drive SD # 1 to DADR4. Drive SD # 2 to Data #
Reading 16 to the cache memory 16 is the same as that at the time of a cache miss in the above-described normal read processing. Further, writing Data # 16 from the cache memory 16 to the drive SD # 1 is the same as the above-described normal writing process.

FIG. 12A shows how the data loss probability changes between the failure recovery processing of the conventional method and the failure recovery processing of this embodiment. In this embodiment,
If the IO status is less than the set value, recover 5D + 1P, 3
Recovery processing is performed in a distributed manner for D + 1P recovery and duplex recovery.

In the state before the recovery of 5D + 1P, 5
For the D + 1P parity group level, 100% data loss will occur no matter which drive other than the failing drive fails at this point. 5
After the recovery of D + 1P, for the parity group level of 3D + 1P, data will not be lost unless either of the two drives involved in 3D + 1P out of the five drives other than the failed drive fails. The probability is 40%.

Furthermore, after the recovery of 3D + 1P, with respect to the parity group level of duplication, no data is lost unless one of the five drives other than the faulty drive that is involved in duplication fails. Therefore, it can be said that the data loss probability is 20%. After the duplex recovery is completed, the recovery process is completed and the data loss probability becomes 0%.

On the other hand, conventionally, for example, a 6D + 1P parity group is set by using all six drives, so that data is lost even if any one of the five drives other than the failed drive fails. Therefore, the probability of data loss is 100% until the entire failed drive is recovered.
Is.

FIG. 12 (b) shows how the performance changes between the fault recovery process of the conventional method and the fault recovery process of this embodiment. In this embodiment, 5D + 1P
The performance gradually recovers with each recovery of 3D + 1P, and recovery of duplex. On the other hand, in the conventional method, the performance remains degraded until the recovery is completed.

Although the data arrangement in the logical group 10 as shown in FIG. 2 has been described in the present embodiment, the data arrangement in the logical group 10 can be freely set at the time of initial setting by the user. Therefore, there are no restrictions. Also, level 5
In, the parity group level can be set to 4D + 2P such that the number of parities in the parity group is increased and the reliability (data loss probability) is further improved.

Further, in the present embodiment, from the viewpoint of reliability (probability of data loss) and performance at the time of failure, parity group levels (5D + 1P, 3D + 1P, and The recovery process was performed in a timely manner in the order of duplexing).

As a modification of this method, there is the following method. First, the higher the importance of the data, the lower the probability of loss is at the parity group level (the loss probability is duplicated, 3
D + 1P, 5D + 1P). The recovery process is performed in the order of the parity group levels with the lowest data loss probability (duplication, 3D + 1P, 5D + 1P). By performing the recovery process that emphasizes only the reliability in this way, the data loss probability for very important data can be greatly reduced.

Further, the following may be carried out. First,
Data to which a read or write request is frequently issued from the CPU 1 is stored in a duplicated parity group that has a small performance deterioration due to a failure and a small data loss probability. On the other hand, data that is not issued such a read or write request has a large performance degradation at the time of failure,
Moreover, the data is stored in the parity group of 5D + 1P, which has a high probability of data loss. Further, when a failure occurs in the drive 12, the recovery processing is performed from the parity group having a high data loss probability. This makes it possible to minimize the period during which performance is degraded due to a failure.

As described above, according to the present invention, it is possible to freely set the recovery processing method depending on the usage environment of the user. Furthermore, it is possible to freely set the time for performing the recovery process and the set value for the status of the read / write request from the upper layer when determining the start of the recovery process. The setting of the recovery processing method is instructed to the MP 20 at the initial setting stage.

Further, as shown below, the present invention can also be used for backup. Drive 12
If the data written on the
A backup is taken by storing the data in the MT or an optical disk. By making a backup in this way, even if a failure occurs in the drive 12 in the disk array and data is lost, it is possible to recover the lost data from this backup data.

By applying the present invention, it is possible to disperse the time for performing this backup processing depending on the characteristics of the parity group level. Specifically, the disappearance probability becomes higher in the order of duplexing, 3D + 1P, and 5D + 1P parity groups. Therefore, a backup is performed more frequently for a parity group with a higher probability of loss. As a result, only the dangerous data with a high probability of data loss need be backed up, so the backup time can be shortened, and the backup time can be distributed according to the characteristics of the parity group level.

Similarly to the failure recovery, the higher the importance of the data is, the lower the data loss probability is stored in the parity group level (the loss probability becomes higher in the order of duplication, 3D + 1P, 5D + 1P), and the backup is made of the data. Order of parity group level with low loss probability (redundancy, 3D
It is obvious that the backup can be performed in the order of + 1P, 5D + 1P) with an emphasis only on reliability.

Similar to the failure recovery process, the backup process can be started when the value becomes equal to or less than the value preset by the user. The backup may be performed by the user by outputting a message prompting the backup from the system, or may be automatically backed up.

Conventionally, backup of all the data in one drive 12 has to be performed, so the backup time is very long, and normal read / write processing must be stopped during backup. I won't. In a system such as a file server used by a large number of users, suspending service continuously for a long time is a big problem as well as failure recovery.
By applying the present invention to backup, it is possible to distribute the backup time according to the characteristics of the parity group level, and therefore the present invention is suitable for use in a file server or the like.

[Embodiment 2] Next, as a second embodiment,
An example of applying the present invention to RAID level 3 will be shown. The configuration and processing procedure of the disk array of this embodiment are the same as those of the first embodiment, so the first embodiment will be described below.
Differences from the embodiment will be described.

FIG. 13 shows the logical group 10 in this embodiment.
An example of the arrangement of the parity groups in the above is shown. In the logical group 10, the SCSI Addr 27 is DADR
1, 2 and 3 are parity groups at a parity group level of 5D + 1P, and Addr 27 in SCSI is DADR.
4, 5 and 6 are parity groups at the 2D + 1P parity group level.

In the present embodiment, the data loss probability is 5D +.
1P is higher than 2D + 1P. Also, at level 3,
In both 5D + 1P and 2D + 1P, the performance does not change between the time of failure and the normal time.

A parity group at the parity group level of 5D + 1P will be described by taking Data # 1 of DADR1 as an example. In the 5D + 1P parity group, Data # 1 which is one piece of data transferred from the upper layer is used as five sub-data (Data # 1-1, Data # 1-
2, Data # 1-3, Data # 1-4, Data #
1-5) and sub-data of each of them is SD #
Write in parallel to the five drives 1, 2, 3, 4, and 5. At this time, from these sub-data, Example 1
Similarly, the parity is created as shown in FIG. 5, and the data is written to the drive SD # 6 at the same time as the writing of the data.

In the parity group of the 2D + 1P parity group level, one piece of data transferred from the upper level is divided into two pieces of sub data, and the respective sub data are written in parallel to the two drives 12. . At this time, a parity is created from these sub data and written at the same time as the data writing.

In the present embodiment, the method of setting the parity group level as described above is the same as in the first embodiment.
The user performs the initial setting for the address table as shown in FIG. Further, address conversion is also performed using the address table shown in FIG. 14 as in the first embodiment.

The address table of FIG. 14 used in this embodiment is almost the same as the address table of FIG. 4, but since this embodiment is RAID level 3, a sub data name field 28 is added.

In this embodiment, the processing of updating the parity when writing data by new or update is the first embodiment.
Is different from. In the first embodiment, level 5 parity update processing is performed, but in the present embodiment, level 3 parity update processing is performed. In the level 3 parity update process, the reading of old data and old parity is not required as in level 5, and the parity as shown in FIG. 5 can be created from the data to be written. Otherwise, the writing process is performed according to the address table shown in FIG. 14 by the procedure shown in the first embodiment.

On the other hand, in writing at the time of a failure, only the parity was updated at the level 5 shown in the first embodiment, but at the level 3 of the present embodiment, in the parity group other than the drive in which the failure has occurred. Sub-data and parity are written in parallel to the normal drive in the same way as in the normal operation.

Further, in the read processing at the time of failure, as in level 5 of the first embodiment, in the parity group, sub-data and parity are read from normal drives other than the drive in which the failure has occurred, and the failed drive is read from these. Restore the sub data stored in. At level 3, the sub data restored in this way and the sub data read from the normal drive are combined, and the CPU
Transfer to 1.

The failure recovery processing when a failure occurs in any of the drives 12 constituting the logical group 10 is stored in the failed drive 12 as shown in FIG. 5 as in the first embodiment. Recovery processing is performed by restoring the sub data from the data and parity stored in the remaining normal drive 12 and storing it in the replaced normal drive 12 or spare drive 12. As in the case of the first embodiment, the recovery processing method sequentially performs recovery processing from a parity group having a low parity group level (high probability of data loss (data loss probability) when a second failure occurs).

As a result, the reliability and the performance at the time of failure can be suppressed to the minimum, and the recovery processing can be performed in a dispersed manner without concentrating.

As a modification of the present embodiment, the first embodiment will be described.
Similar to the modification example, the higher the importance of data, the lower the probability of loss may be stored in the parity group level, and the recovery process may be performed in the order of the parity group level in which the probability of data loss is lower. By performing the recovery process that emphasizes only the reliability in this way, the data loss probability for very important data can be greatly reduced.

As described above, in the present embodiment and the first embodiment, the parity group forming the logical group 10 is
The same effect as that of the first embodiment is obtained only by the difference between level 5 and level 3. From this, in the configuration of the present embodiment,
It is naturally possible to apply RAID level 5.

[Third Embodiment] The first embodiment has shown the method of freely setting the parity group level in the parity group forming the logical group based on the level 5. Further, in the second embodiment, the method of freely setting the parity group level in the parity group forming the logical group based on the level 3 has been described.
By doing so, it is possible to minimize the reliability and the performance in the event of a failure, and to distribute the recovery processing time.

Therefore, in this embodiment, an example will be described in which different RAID levels are set in the parity groups forming the logical group. The user can set different RAID levels when setting the parity group level on the address table at the initial setting stage. Hereinafter, points different from Example 1 will be shown.

FIG. 15 shows the logical group 10 in this embodiment.
An example of the arrangement of the parity groups in the above is shown. In the logical group 10, Addr27 in SCSI is DADR1
Is a parity group of a parity group level of 5D + 1P in RAID level 3, and is Add in SCSI.
r27 is a DADR 2 and 3 is a parity group of a parity group level of 5D + 1P in the RAID level 5, and Addr 27 in SCSI is a DADR 4 is a RAID group
3D + 1P parity group level and redundant parity groups coexist at Level 3 of the above, and Addr 27 in SCSI is DADR 5,6.
In RAID level 5, a parity group of a 3D + 1P parity group level and a redundant parity group are mixed.

FIG. 16 shows the address table at this time. The address table is almost the same as that of FIG. 4, but since different RAID levels are mixed in this embodiment, R
An AID level field 29 is provided. Further, a sub data name field 28 for use in the case of RAID level 3 is provided.

In this embodiment, as described above, RAID
The method of setting the level and the parity group level is carried out by the user in the initial setting for the address table as shown in FIG. 16, as in the first and second embodiments.
In addition, address conversion is performed in the same manner as in the first and second embodiments.
The address table shown in FIG.

The data writing method (parity creating method) and reading method in the parity group of 5D + 1P parity group level in the RAID level 3 of DADR1 and the parity group of the parity group level of 3D + 1P in DADR4 are Both are the same as in the second embodiment.

On the other hand, the data writing method (parity creating method) and the reading method in the parity group of the parity group level of 5D + 1P in the RAID level 5 of DADR2 and 3 and the parity group of the parity group level of 3D + 1P in DADR5 and 6 are as follows: The same as in the first embodiment for both normal and fault conditions.

The method of writing data (parity creating method) and the method of reading data in the parity group at the parity group level of the dual DADRs 2, 5, and 6 are the same as those in the first embodiment both in the normal state and in the fault state.

In this embodiment, the failure recovery processing when a failure occurs in any of the drives 12 constituting the logical group 10 is the same as in the second embodiment and RAI in the RAID level 3.
At level 5 of D, the same as in the first embodiment, both of which perform sequential recovery processing from a parity group having a low parity group level (high probability of data loss when a second failure occurs (data loss probability)). To do. As a result, also in the present embodiment, as in the first and second embodiments, it is possible to suppress the reliability and the performance at the time of failure to the minimum and perform the recovery processing in a distributed manner instead of concentrating it. Become.

In this embodiment, there are a plurality of RAID levels and a plurality of parity group levels for each RAID level in one logical group. When the user has a large amount of data and wants to perform high-speed transfer between the CPU 1 and the disk array to shorten the data writing / reading time, the user gives an instruction to write in the RAID level 3 area. At this time, the parity group level is selected to be 5D + 1P or 3D + 1P depending on the amount of data, the required transfer rate, and the reliability. That is, when the data amount is large and high-speed transfer is required, the parity group level is designated as 5D + 1P, and when relatively reliable is required, the parity group level is set as 3D + 1P.

[0206] Further, since the user has a small amount of data, if high-speed transfer at RAID level 3 has no effect, the user instructs to write in the area of RAID level 5. Further, the user gives an instruction to write in a duplicated area when the amount of data of the user is small, frequent write requests occur, and important data.

As described above, the request instructed by the user via the CPU 1 is checked in the address table in the MP 20 of the ADC 2 and the RAID is applied to the requested write area.
It recognizes the level 3 and level 5 of the above, the vacant area for duplication or the area to be updated, and controls to write to the area.

As described above, this embodiment and the first embodiment,
With No. 2, the effects of Examples 1 and 2 can be obtained only by mixing RAID levels 5 and 3 and duplication (level 1) in the parity group forming the logical group 10.

[0209]

In the conventional disk array, all parity group levels forming a logical group are unified as 5D + 1P as shown in FIG.
For this reason, the probability of data loss becomes uniform, and it has been necessary to stop for a long time in order to recover from failure in order to secure reliability. If a logical group is configured with a single parity group level as in the conventional case, and recovery processing of a drive in which a failure has occurred is performed by dividing it as in the above embodiment,
The data loss probability and the performance at the time of failure remain unchanged until the recovery processing is completed, and the longer the recovery processing is completed, the higher the probability of data loss.

However, according to the present invention, a logical group is formed by a plurality of parity group-level parity groups having different reliability and performance, and data recovery processing of a drive in which a failure has occurred is performed in the ascending order of parity group level (data Order of decreasing probability of loss and performance in case of failure). As a result, reliability is improved each time the recovery process is performed in units of parity group levels, and thus the probability of data loss is low even if the recovery process is completed for a long time.
Moreover, the performance at the time of failure is improved every time the recovery process is performed in units of parity group levels.

From the above, according to the present invention, it is possible to disperse the time for performing the recovery process, and it is possible to perform the recovery process efficiently.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram of a first embodiment.

FIG. 2 is a data layout diagram in a logical group according to the first embodiment.

FIG. 3 is a data layout diagram in a logical group of a conventional disk array

FIG. 4 is an explanatory diagram of an address table according to the first embodiment.

[Figure 5] Parity creation explanatory diagram

FIG. 6 is a flowchart of writing processing of new data.

FIG. 7 is a flowchart of update writing processing.

FIG. 8 is a flowchart of update writing processing at the time of failure.

FIG. 9 is a flowchart of a reading process.

FIG. 10 is a flowchart of read processing at the time of failure.

FIG. 11 is a flowchart of failure recovery processing.

FIG. 12 is an explanatory diagram of a conventional method and failure recovery processing of the present invention.

FIG. 13 is a data layout diagram in a logical group according to the second embodiment.

FIG. 14 is an explanatory diagram of an address table according to the second embodiment.

FIG. 15 is a data layout diagram in the logical group according to the third embodiment.

FIG. 16 is an explanatory diagram of an address table according to the third embodiment.

[Fig. 17] Internal view of the drive

FIG. 18 is an explanatory diagram of allocation of parity groups.

[Explanation of symbols]

1: CPU, 2: Array disk controller (AD
C), 3: channel path director, 4: external interface path, 5: interface adapter, 6: channel path switch, 7: cluster, 8: channel path, 9: drive path, 10: logical group, 12: drive, 13 : Channel interface (CH I
F) Circuit type, 14: Data control circuit (DCC), 15:
Channel side cache adapter (C Adp), 16:
Cache memory, 17: Drive side cache adapter (C Adp), 18: Drive interface circuit (Drive IF), 19: Parity generation circuit (P
G), 20: microprocessor (MP), 21: parity group level, 22: data name, 23: cache address, 24: data drive number (DDDrive)
No. ), 25: failure flag, 26: parity drive number (PDrive No.), 27: address in SCSI drive (Addr in SCSI), 28: sub data name, 29: RAID level.

Claims (31)

[Claims] 1. A disk array device comprising a disk device including a plurality of drives forming a logical group and a controller for managing the disk device, wherein i (i is an integer of i ≧ 1) pieces of data. And an area for storing a first parity group composed of j (j is an integer of j ≧ 1) error correction data created from the data, and a configuration different from the first parity group And an area for storing the second parity group of
A disk array device mixed in the logical group. 2. A disk array device comprising a disk device including n drives forming a logical group, and a control device for managing the disk device, wherein each drive in the logical group is partitioned into n partitions. Select any number of arbitrary partitions of m or less arbitrary drives, and set a partition group by the selected partitions, and the logical group is configured by a plurality of partition groups in which m is different from each other. Characteristic disk array device. 3. A disk array device comprising a disk device storing or storing the data in response to an input / output request of the data from a host device, and a control device managing the disk device. , The disk device is composed of a large number of drives, these drives are grouped into logical groups of two or more n drives, and each drive in each logical group is divided into partitions, and n or less m A disk characterized by selecting an arbitrary number of arbitrary partitions of arbitrary drives and setting a partition group by the selected partitions, wherein the logical group is composed of a plurality of partition groups having mutually different m. Array device. 4. The disk array device according to claim 2, wherein when the partition group is set in the logical group, the addresses of the selected partitions are the same in each drive. 5. The disk array device according to claim 2, wherein when the partition group is set in the logical group, a partition group of a type and size according to a user's designation can be set. 6. The disk array device according to claim 2, wherein when the partition group is set in the logical group, a partition selected from each drive is a set of cylinders. 7. The disk array device according to claim 2, wherein when the partition group is set in the logical group, the partition selected from each drive is set in track or record units. 8. The disk array device according to claim 2, wherein the partition group set in the logical group is managed by a table in the control device. 9. The disk array device according to claim 2, wherein a RAID is set for each partition group set in the logical group. 10. At least two partition groups set in the logical group have different RAs from each other.
10. The data of a parity group, which is configured at an ID level, and stores in each partition group data of a plurality of data and one or more error correction data created from the data. Disk array device. 11. The disk according to claim 9, wherein at least two partition groups set in the logical group are configured with the same RAID level in which the number of the error correction data is different. Array device. 12. The disk array device according to claim 9, wherein at least two partition groups set in the logical group are configured with the same RAID level having different performances. 13. When setting two or more partition groups in the logical group, different RAID levels set for each partition group are classified as parity group levels, and a table corresponding to each parity group level is set. 10. The disk array device according to claim 9, wherein each partition group is managed by. 14. A failure occurs in any drive in the logical group, and the data in the failed drive is recovered from the data stored in the normal drive in the logical group and the error correction data. 14. The disk array device according to claim 13, wherein the failure recovery processing is performed for each parity group level in the failure recovery processing. 15. When performing failure recovery processing for each parity group level, normal read and write processing is performed for a certain period of time after failure recovery processing of a certain parity group is completed, and then failure of the next parity group. 15. The disk array device according to claim 14, wherein failure recovery processing for recovering data in a failed drive is repeated by repeating recovery processing. 16. An arbitrary drive in the logical group fails, and data in the failed drive is recovered from data stored in a normal drive in the logical group and error correction data. 16. The disk array device according to claim 14, wherein the failure recovery processing is performed sequentially from a parity group level with low reliability and low data loss probability. 17. A failure occurs in any drive in the logical group, and the data in the failed drive is recovered from the data stored in the normal drive in the logical group and the error correction data. 16. The disk array device according to claim 14, wherein the failure recovery processing is performed sequentially from the parity group level with high reliability and high data loss probability. 18. The disk array according to claim 10, wherein data requiring high reliability is stored in a parity group in which a RAID level with high reliability and high data loss probability is set. apparatus. 19. A failure occurs in any drive in the logical group, and the data in the failed drive is recovered from the data stored in the normal drive in the logical group and the error correction data. 16. The disk array device according to claim 14, wherein the failure recovery processing is performed from the parity group level, which has a large decrease in performance at the time of failure. 20. A failure occurs in any drive in the logical group, and the data in the failed drive is recovered from the data stored in the normal drive in the logical group and the error correction data. 16. The disk array device according to claim 14, wherein the failure recovery processing is performed from the parity group level in which the performance degradation at the time of failure is small. 21. Data that is frequently issued by a higher-level device for read or write requests is stored in a parity group in which a RAID level that causes a large decrease in performance at the time of failure is set, and failure recovery processing is immediately performed when a failure occurs. 11. The method according to claim 10, wherein the normal state is restored.
9. The disk array device according to item 9. 22. Data which is frequently issued by a host device for read or write is stored in a parity group in which a RAID level is set so that performance degradation at the time of failure is small, and failure recovery processing is immediately performed when a failure occurs. The method according to claim 10, wherein the normal state is restored.
0. The disk array device according to item 0. 23. The disk array device according to claim 13, wherein when the data in the logical group is backed up, a backup process is performed for each parity group level. 24. When data in the logical group is backed up for each parity group level, normal read and write processing is performed for a certain time after the backup processing of a certain parity group is completed, and then the next 24. The disk array device according to claim 23, wherein the backup processing of the parity group is repeated to back up the data in all the drives forming the logical group. 25. When performing backup for each parity group level on the data in the logical group, the backup process is sequentially performed from the parity group level having the low reliability and the low data loss probability. 24. The disk array device according to item 24. 26. When the data in the logical group is backed up for each parity group level, the backup process is sequentially performed from the parity group level with high reliability and high data loss probability. 24. The disk array device according to item 24. 27. The failure recovery process of the parity group is started when the number of read or write requests from the host device becomes smaller than a value preset by the user. Disk array device. 28. The disk according to claim 23, wherein when the number of read or write requests from the host device becomes smaller than a value preset by the user, the backup processing of the parity group is started. Array device. 29. When the data in the logical group is backed up for each parity group level, the backup process is performed more frequently for a parity group having a higher data loss probability.
The disk array device according to 1. 30. A method of dividing an area of a disk device in a disk array device comprising a disk device including a plurality of drives forming a logical group, and a control device for managing the disk device, wherein i ( i is an integer of i ≧ 1) and an area for storing a first parity group composed of j (j is an integer of j ≧ 1) error correction data created from the data , An area for storing a second parity group having a configuration different from that of the first parity group,
A method of partitioning a disk array, wherein the areas of the disk devices are partitioned so that they are mixed in the logical group. 31. A method of partitioning an area of a disk device in a disk array device comprising a disk device including n drives forming a logical group, and a controller for managing the disk device, Each drive in the group is divided by a partition, any number of n or less m arbitrary drives are selected, a partition group is set by the selected partition, and the logical groups are different from each other in m. A disk array partitioning method characterized by partitioning into a plurality of partition groups.