JP3453376B2 - Mediation method, mediation device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a redundant configuration type data array storage for storing continuous media data such as video information.
The present invention relates to an arbitration method or an arbitration device in a storage device .

[0002]

2. Description of the Related Art With the development of multimedia technology in recent years, continuous media data such as digitized video / audio data is often recorded in a randomly accessible data storage device such as a hard disk drive. It's starting. Further, since the continuous media data has a large data size, an array server system is constructed by using a plurality of data storage devices, and the server system is virtually recognized as one large-capacity device to the outside. It is becoming common to use

For such a server system, normally, a block of data having a large size is continuously accessed and transferred at a high data transfer rate. When the data to be recorded is video data, for example, if it is of the MPEG standard, the transfer rate is about 1.5 Mbps to 30 Mbps.

Further, in accessing continuous media data, not only a high data transfer rate, but also a certain response performance is required so as not to be interrupted in the reproduction of a moving image / voice. Furthermore, even if a part of the data storage device fails, the data storage array device is required to have reliability that does not stop. At this time, RAID (R
edandant Array of Inexpens
If a plurality of data storage devices are configured in a redundant array using the IVE Disks technology, even if a part of the data storage device fails, the correct data can be restored by the redundant data.

FIG. 1 shows the configuration of a server system composed of a RAID 3 type data storage array device. In the data storage array device, the data storage devices 4a to 4d
Is a data storage device for recording data. The data storage device 4e is a data storage device that records redundant data. The external host device 1 is means for issuing a data access request to the data storage array device. Between the external host device 1 and the data storage array device, for example,
FastWide SCSI standard interface and F
It is connected by an interface of the ibreChannel standard. The controller 3 is means for receiving a data access request issued from the external host device 1, dividing the data access request to the data storage devices 4a to 4e and issuing the data access request, and performing data access to the data storage devices 4a to 4e. For example, in the case of a read request, the data in the data storage devices 4a to 4e is read and transferred to the host device 1. When the access request to the data storage devices 4a to 4e is completed, the transfer completion notification is sent to the external host device 1.

Next, regarding the method of writing and reading data in the RAID 3 type data storage array device,
This will be described with reference to FIG. According to the data access request from the external host device 1, the controller 3 manages the data and the redundant data as a cluster as a cluster. As shown in FIG. 2, five data storage devices 4a-
In the case of 4e, a cluster of cluster data is 5 Di
It is divided into sk and stored. This cluster data is 4
Data itself is divided and stored in each of the data storage devices 4a to 4d, and redundant data is stored in the data storage device 4e. A parity method is generally used as a method of generating redundant data. The data storage device 4e is a data storage device for parity, and includes four data storage devices 4e.
Parities calculated from the data stored in a to 4d are stored.

When writing data from the external host device 1, the controller 3 calculates a parity value for the data sent from the host and then stores the parity value in the five data storage devices 4a to 4e. When a data read request is received from the external host device, normally, data from four normal data storage devices 4a to 4d are collectively transferred to the external host device. Data storage device 4a-
When a part of 4d fails, the data storage device 4
The data can be restored by performing the parity calculation from the normal three of a to 4d and the data read from the parity data storage device 4e, and the restored data is transferred to the host device 1.

Next, the operation of the data storage device after replacement will be described when a failure occurs in the data storage device.
In the data storage array device using the RAID technology, as can be seen from the read operation described above, even if a part of the data storage device fails, the data can be restored without stopping the system. Is. When a part of the data storage device fails, a state where the failed data storage device is not used and access is continued without redundancy is called a degenerate operation.

However, since there is no redundancy during this degeneracy operation, it is necessary to replace the faulty data storage device and restore the normal state. When the failed data storage device is replaced, it is necessary to reconstruct the data from the remaining normal data storage device and restore the data to the replaced data storage device. This operation is generally called a reconstruction operation.

The reconstruction operation is executed in the steps of reading data from the data storage device, performing a parity operation, and writing the regenerated data. For example, when the data storage device 4a is replaced, the reconfiguration operation is performed for each of all the clusters that need to be reconfigured.
4e is read and a parity operation is performed to generate data in the data storage device 4a and store it in the data storage device 4a. The series of reconfiguration operations are repeated until there are no redundant clusters. Also, this reconstruction process
It is executed in parallel with the external access operation with the external host device 1, and the system can be operated without stopping. In this way, a highly reliable storage device can be realized.

[0011]

However, the conventional data storage array device has the following problems. As shown in the related art, in the data storage array device in the reconfigured state, the external access request operation comes during the reconfiguring operation. In this case, the external access request operation is performed after the reconfiguration operation is completed. Since the access operation is performed, the external access performance is deteriorated. As a technique for solving such a problem, for example, Japanese Patent Laid-Open No.
JP-A-127839 discloses a technique in which the number of reconfiguration operations is reduced by controlling the operation frequency of the reconfiguration operations in accordance with the frequency of external access requests, and external access is performed during that period.

Further, Japanese Patent Laid-Open No. 10-161817 discloses a technique for operating the external access request operation with a higher priority than the reconfiguration operation. For example, if there is a data write request from the outside during the reconfiguration operation and the data storage device to be written is the same as the data storage device currently the reconfiguration operation target, The write request is given priority and the reconfiguration operation is suspended. The inventor estimates that the reconstruction operation is not executed after the data writing is completed.

However, as described above,
For a data storage array device that handles continuous media data such as voice, not only average access performance but also high-speed response performance to external access is required. The above-described two disclosed examples contribute to the improvement of the average external access performance band, but cannot improve the external access response performance of the data storage device.

This is because in the conventional reconstruction processing including these two disclosed examples, a series of three operations including at least a read operation to the data storage device, a data restoration operation, and a write operation to the data storage device are performed. Since the operations are inseparable (cannot be separated), if an external access from the host device occurs while the data storage array device is in the process of being reconfigured, defer the external access until the reconfiguration operation ends. Is inevitable,
The access response performance to the external access request could not be improved.

[0015] The present invention has been made to solve the above problems, while ensuring data access performance from the outside of the continuous media data, the arbitration method and arbitration reconstruction process when redundancy is enabled The purpose is to provide a device and the like.

[0016]

In order to achieve the above object, the first invention (corresponding to claim 1) includes a plurality of segments obtained by dividing a recording area of a plurality of data storage devices. , A plurality of parity groups in which the segments of each of the plurality of data storage devices are grouped using a predetermined algorithm, and redundant data is stored in at least one of the segments in the plurality of parity groups. In a data storage array device, when an access from an external host device occurs during the reconfiguration process, an arbitration method for arbitrating with an access required for the reconfiguration process, the external host device being provided during the reconfiguration process. there is a write access request different parity group f is a read access request, or parity group during the reconstruction process from When the reconfiguration processing is in the reading step or the reconfiguration step in the step determination step for determining the step in which the reconfiguration processing is in progress, and the external host device Read access or write access from the device before the writing step in the reconfiguration processing, and the reconfiguration processing is performed by the arbitration step.
From the external host device at the same time as the reconstruction process in
It is an arbitration method characterized by enabling the access of .

[0017]

The second invention (corresponding to claim 2 )
Has a plurality of segments formed by dividing a recording area of a plurality of data storage devices, and a plurality of parity groups obtained by grouping the segments of each of the plurality of data storage devices using a predetermined algorithm. However, when an access from an external host device occurs during the reconfiguration process of the data storage array device that stores redundant data in at least one of the segments in the plurality of parity groups, the access necessary for the reconfiguration process is performed. An arbitration device that performs arbitration of the reconfiguration process when a read access request from an external host device during the reconfiguration process or a write access request to a parity group different from the parity group under the reconfiguration process is issued. The process determination means for determining the process in which the configuration process is in progress, and the reconfiguration by the process determination means. Processing the read process, or when it is determined that the reconfiguration process, a read access or a write access from the external host device, said by interrupting before the writing process in reconstruction processing and a mediation unit, The reconstruction processing is performed by the arbitration means.
In parallel with the reconfiguration process in
It is an arbitration device that can be accessed from .

[0019]

The third invention (corresponding to claim 3 )
In the arbitration method of the first aspect of the present invention, when there is a read access request from the external host device during the reconfiguration processing or a write access request to a parity group different from the parity group under the reconfiguration processing, In the step determination step for determining the step in which the reconfiguration processing is in progress, and in the step determination step, when it is determined that the reconfiguration processing is in the reading step or the reconfiguration step,
A program for causing a computer to execute an arbitration step of interrupting a read access or a write access from the external host device before a writing step in the reconfiguration processing.

A fourth invention (corresponding to claim 4 )
Is a recording medium that carries the program of the third aspect of the present invention and is a recording medium that can be processed by a computer.

[0022]

[0023]

[0024]

[0025]

[0026]

[0027]

[0028]

[0029]

[0030]

[0031]

[0032]

[0033]

[0034]

[0035]

[0036]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described in detail below with reference to the drawings.

(Embodiment 1) The configuration of a data storage array device in which the online reconfiguration processing device according to the first embodiment of the present invention is used is basically the same as that of the conventional example. The description of common points with the conventional example will be omitted.

An example of the configuration of the controller 3 in the data storage array device will be described with reference to FIGS. 3 and 4. The host execution unit 31 is means for controlling analysis / completion notification of a data access request from the external host device 1 and access data transfer. The array control unit 33 is means for converting a data access request into a cluster access request. The redundancy calculator 34 is means for generating redundant data in the cluster data and restoring the cluster data generated due to a failure of the data storage device. The data storage device controller 35 uses the data storage device 4 based on the cluster access request issued from the array controller 33.
It is a means for controlling a to 4e. Cluster control unit 32
Refers to the cluster management table 36 and restores the cluster data whose redundancy is lost (a part of the cluster data is lost due to the failure or replacement of the data storage device) to the redundancy state again. is there. The cluster management table 36 is a table held in the cluster control unit 32 for managing the redundant state of cluster data.

With respect to the online reconfiguration processing apparatus according to the present embodiment having the above-mentioned configuration, a data management method and basic operations for creating / restoring redundant data, writing / reading data and reconfiguration will be described. At the same time, an embodiment of the arbitration method of the present invention will be described.

First, the data management method in the data storage array apparatus 2 and the redundant data creation and data restoration processing will be described with reference to FIG. The controller 3 collects a data group associated with a data access request from the external host device 1 and redundant data calculated using this data group as one group (corresponding to each other) and manages them as cluster data. In the case of five data storage devices 4a-4e, this cluster data is four data storage devices 4a-4d.
The data itself is divided and stored in, and the redundant data is stored in the remaining one data storage device 4e. A parity method is generally adopted as a method of generating redundant data. For example, when writing data from the external host device 1, the parity value is calculated and then stored in the five data storage devices 4a to 4e. The parity that is redundant data is set such that the total number of 1s of the data values of the four data storage devices 4a to 4d is an even number including the parity bit. When the data value is 0, 1, 0, 0, the parity value is 1, and this value is stored in the parity data storage device 4e. When a data read request is received from the external host device 1, cluster data is read from the five data storage devices 4a to 4e. At this time,
When a failure occurs in one of the data storage devices 4a to 4d, if a part of the cluster data is read from the remaining four data storage devices, parity calculation is performed, and the cluster lost due to the failure is read. You can restore some of the data. For example, in the above case, it can be understood from the values 0, 1, 0 of the data storage devices 4a to 4d and the parity value 1 of the data storage device 4e that the data of the remaining one data storage device 4e is 0. Although the even parity is used here, the same configuration can be made by using the odd parity.

Next, the data write operation in the data storage array device will be described. Host execution unit 31
When the data write request is received from the external host device 1, the data write request is notified to the array controller 33. The array control unit 33 converts the data write request into a cluster access request, uses the redundancy calculation unit 34 to calculate redundant data (parity data) corresponding to the write data, and creates cluster data. Then, via the data storage device control unit 35, the data storage devices 4a-4
Write cluster data to e. When the writing of the cluster data is completed, the host execution unit 31 notifies the external host device 1 that the data writing request has been completed.

Next, a data read operation in the data storage array device 2 will be described. Host execution unit 3
When 1 receives a data read request from the external host device 1, it notifies the array control unit 33 of this data read request. The array control unit 33 converts the data read request into a cluster access request, and reads the cluster data from the data storage devices 4a to 4e via the data storage device control unit 35. When the reading of the cluster data is completed, the array control unit 33 checks whether or not a part of the cluster data is missing due to a failure or replacement of the data storage device.

At this time, if a part of the cluster data is missing due to a failure or replacement of the data storage device, a part of the missing cluster data is restored from the remaining cluster data by using the redundancy calculator 34. To do. The host execution unit 31 notifies the external host device 1 that the transfer of read data and the data read request have been completed.

Next, the state transition of reconfiguration in the data storage array device 2 will be described with reference to FIG. As can be seen from the above read operation, even if a part of the data storage device fails, the data can be restored without stopping the system. In this way, in the event of a data storage device failure, the state in which access is continued without redundancy without using the failed data storage device,
This is called a degenerate operation. Since there is no redundancy during this degeneration operation, it is necessary to replace the failed data storage device and restore the normal state. When the failed data storage device is replaced, it is necessary to reconstruct the data from the remaining normal data storage device and restore the data to the replaced data storage device. This operation is called reconfiguration operation. As shown in FIG. 5, the normal state indicated by "normal access" changes to the "degenerate" state when a failure occurs, and the "reconfiguration" state after replacement of the failed data storage device.

Here, the reconfiguration operation in the data storage array device 2 will be described. The cluster control unit 32
The cluster management table 36 is referred to search for a cluster without redundancy. When the cluster control unit 32 detects a cluster having no redundancy, the cluster control unit 32 notifies the array control unit 33 of a read request for the cluster. The array control unit 33 transmits the cluster data to the data storage devices 4a to 4e via the data storage device control unit 35 as in the external data read operation.
To read the cluster data. When the reading of the cluster data is completed, the array control unit 33 restores the cluster data corresponding to the exchanged data storage device from the remaining cluster data using the redundancy calculation unit 34. The array control unit 33 that has completed the restoration of the cluster data notifies the cluster control unit 32 that the cluster read request has been completed.

Next, the cluster control unit 32 notifies the array control unit 33 of a write request for the cluster data read out earlier. The array control unit 33 writes the cluster data to the data storage devices 4a to 4e via the data storage device control unit 35 as in the external data writing operation. When the writing of the cluster data is completed, the array control unit 33 notifies the cluster control unit 32 that the data write request is completed. The cluster control unit 32 records in the cluster management table 36 that the redundancy of the cluster for which the reconfiguration processing has been completed has returned.

The series of reconfiguration operations are repeated until there are no clusters without redundancy based on the cluster management table 36. Further, this reconfiguration operation is executed in parallel with the external access operation with the external host device 1,
The system can be operated without stopping.

Next, the operation of this embodiment will be described in detail with reference to FIGS. 6 and 12 to 16. 15 and 16
6 is a flowchart for the cluster control unit 32 to manage the cluster access right for access to the cluster. 12 and 13 show read (external Read) and write (external Write) operations of the data storage array apparatus 2 in response to an access from the host apparatus 1. FIG. 14 is a flowchart of the reconfiguration operation in the cluster control unit 32. External Read and External Write
When the cluster is accessed in the reconfiguration operation, the access right is checked for each cluster and the actual access is executed.

Further, the cluster management table 3 shown in FIG.
In the control state of the access state for each cluster of 6,
Reconstruction read 52, data reconstruction 55, reconstruction wri
When an external access is made during the operation of te53, the state is shifted to the access state 54 and the access state 5
4 is continued until the write operation is completed. The number of times the host execution unit 31 has accessed the cluster is “write” in the cluster management table 36.
"count", and the access status 54
At, when a predetermined “write count” is performed, the access state of the cluster returns to the standby state 51. The access state for each cluster is executed and managed by an access right acquisition routine and an access right release routine in the controller. In this embodiment, the acquisition and release of the access right can be implemented as a part of the function of the array control unit 33, but may be configured independently of the array control unit 33.

Next, with reference to FIG. 12, the operation when a data read request is made from the outside will be described. Upon receiving a data read request from the outside, the host execution unit 31 acquires a cluster access right (S901).
As a result, the host execution unit 31 accesses the array control unit 33 to read the data in the data storage devices 4a to 4e (S902). When the data in the data storage devices 4a to 4e is read, the host execution unit 31 releases the cluster access right (S903). Next, the host execution unit 3
1 determines whether or not there is a data error in the read data (S904), and if there is a data error, restores the read data using the redundancy calculation unit 34 (S9).
05), if not present, the data is sent to the outside as it is, and the read operation is completed.

Next, with reference to FIG. 13, an operation when a data write request is made from the outside will be described. Upon receiving a data write request from the outside, the host execution unit 31 acquires a cluster access right (S100).
1). After that, the redundancy calculation unit 34 generates the parity data stored in the data storage device 4e from the data to be written (S1002). After this, the array controller 3
3 writes the actual data to the data storage devices 4a to 4d and the parity data to the data storage device 4e (S1003). When the writing is completed, the host execution unit 31 releases the class access right (S1004),
The write operation ends.

Next, the reconfiguration operation of the online reconfiguration processing apparatus according to this embodiment will be described with reference to FIG. First, the cluster control unit 32 determines whether or not a cluster requiring reconfiguration exists in the data storage devices 4a to 4e according to the cluster management table 36 (S1101), and when there is a cluster requiring reconfiguration. Determines the parity group to which the cluster belongs (S1102). Clusters that need to be reconfigured include those that have not been reconfigured after the replacement of the data storage device, or those that have incomplete data due to a temporary failure.

Next, the cluster access right for the determined parity group is acquired (S1103), and the array control unit 33 reads data from the data storage devices 4a-4e (Read) (S1104). Next, the redundancy calculation unit 34 reconstructs the data from the read data and the redundant data (S1105). Next, the cluster access right is checked (S1106), and when it is determined that Write to the cluster is necessary,
The reconfigured data is written, and when the reconfigured data is completely written, the access right of the reconfigured cluster is released (S1108). Here, data access from the outside can be received at any place in steps S1101 to S1108.

Here, the operation when there is a data access from the outside during S1104 to S1108 will be described with reference to FIG.
Will be described with reference to.

FIGS. 17A, 17B and 17C are timing charts for explaining the reconfiguration operation when there is a data read request as an external access request.

First, the control state of the access state of each cluster in the cluster management table 36 will be described with reference to FIG. In the figure, the standby state 51 is a state in which there is no access to the cluster. When the cluster is reconfigured, the state shifts to the reconfiguration read state 52, and when the reconfigured data is written to the disk device, the reconfiguration write is performed.
Transition to state 53. At this time, when an external access is made during the operation of the reconfigured read 52 or the reconfigured write 53, the access state 54 is entered, and the access state 54 is continued until the write operation is completed.

Next, with reference to FIGS. 17A, 17B and 17C, the operation when a data read request is made from the outside during the reconstruction process will be described. 17A shows a case where a read request is issued during execution of the reconstructed read state 52, and FIG. 17B shows a case where a read request is issued during execution of the data reconstructed state 55. ) Is the reconstructed w
The case where a read request is issued during execution of the write state 53 is shown.

In the following operation, it is common whether the data to be externally requested to be read is the same as or not the same as the cluster to be reconstructed.

First, with reference to FIG. 17A, the operation when a read request is issued during execution of the reconfigured read state 52 will be described.

When the operation starts (in the figure, "star
t ”), the controller 3 controls the data storage devices 4a-4e.
In response, a read command is output. When the read command is accepted by the data storage devices 4a to 4e, the data storage devices 4a to 4e read the cluster data to the controller 3 and output the cluster data. Here, a period 1701 is a time until a read command is output and is accepted by the data storage devices 4a to 4e, and in a period 1704, the data storage devices 4a to 4e start reading cluster data from the controller 3. It is the time from the end to the end.

Next, when the controller 3 completes the input of the cluster data from the data storage devices 4a to 4e,
Reconstruction operation is performed using cluster data and parity. Here, the period 1702 is the time from the start to the end of the reconstruction operation.

Next, the controller 3 outputs a reconstructed data write command to the data storage devices 4a to 4e. The write command is the data storage device 4a to 4e.
When received, the data storage devices 4a to 4e receive the reconstructed data from the controller 3 and perform writing. Here, the period 1703 is the time until the write command is output and is accepted by the data storage devices 4a to 4e.

In the above operation, during the period 1704, when the host device 1 issues a data read request 1710 to the data storage array device 2 and the controller 3 accepts it, the data storage devices 4a ... 4e performs the read operation during the period 1704, but the period 17
When 04 ends, the controller 3 requests the read request 17
According to 10, the corresponding data is read from the data storage devices 4a to 4e. Here, the period 1706 is the time from when the data storage devices 4a to 4e start outputting data to the host device 1 to when it ends.

At this time, the reconstruction operation is performed by the controller 3
The data read operation is performed by the redundancy calculation unit 34 in the data storage devices 4a to 4e and the data storage device management unit 3.
5. The array control unit 33 and the host execution unit 31 perform these two operations simultaneously.

Further, if the data reading to the host device 1 is continued even after the reconfiguration of the cluster data by the redundancy calculation unit 34 is finished, the data is read to the host device 1 even if the reconfiguration operation is finished. The write command is not output to the data storage devices 4a to 4e until the data reading is completed. In FIG. 17A, the period 1706
Is the surplus time t1 in addition to the periods 1702 and 1703.
Just have a length.

Next, when the period 1706 ends, the controller 3 outputs the reconstructed data to the data recording devices 4a to 4e, and writes this data. Here, the period 1703 ′ is the time until the write command is output and is accepted by the data storage devices 4 a to 4 e, and its size is equal to the period 1703.

Finally, when the write command is accepted, the data storage devices 4a to 4e respectively write the reconstructed data output from the controller 3, and the process is completed. Here, the period 1705 'is the data storage device 4a.
4e is the time from the start to the end of writing the data output from the controller 3.

When there is no read request from the outside, the write command from the controller 3 is accepted immediately after the end of the period 1702, and the data storage devices 4a to 4e write the data output from the controller 3. Here, the period 1705 is the time from when the data storage devices 4 a to 4 e start writing the data output from the controller 3 to when the data storage devices 4 a to 4 e finish writing, and the magnitude thereof is the period 17
Equal to 05 '.

Comparing the case where there is no read request from the outside with the case where there is a read request from the outside, the reconfiguration operation ends in the period 170 ′ (period 1703) when the read request is received from the outside as compared to when there is no read request. ) + Excessive time t1.

However, the period 1706 of the reconstruction operation
Since the data reading 1703 from the outside and the data reading 1703 from the outside are performed in parallel, the time for the reconfiguration operation when there is data reading from the outside is significantly shortened as a whole.

Next, with reference to FIG. 17B, the operation when a read request is issued during execution of the data reconstruction state 55 will be described.

From the start of the operation, the data storage device 4a
Up to the reconstruction operation of the cluster data read from 4e to 4e is the same as that in FIG. Here 1702
Is the time from the start to the end of the reconstruction operation.

Here, during the period 1702, when the host device 1 issues a data read request 1710 to the data storage array device 2 and the data is accepted by the controller 3, the controller 3 sends a read request 1710. In response, the corresponding data is read from the data storage devices 4a to 4e. At this time, the reconfiguration operation is performed by the redundancy calculation unit 34 in the controller 3, and the data read operation is performed by the data storage devices 4a to 4e, the data storage device management unit 35, the array control unit 33, and the host execution unit 31. These two operations are performed in parallel at the same time.

As in the case of FIG. 17A, the redundancy calculator 3
If the data reading to the host device 1 is continued even after the reconfiguration of the cluster data by 4 is completed, even if the reconfiguration operation is finished, the data reading to the host device 1 is finished. No write command is output to the data storage devices 4a to 4e. In FIG. 17B, the period 1706 has a length of the surplus time t2 in addition to the periods 1702 and 1703.

Next, when the period 1706 ends, the controller 3 outputs the reconstructed data to the data recording devices 4a to 4e, and causes this data to be written. Here, the period 1703 ′ is the time until the write command is output and is accepted by the data storage devices 4 a to 4 e, and its size is equal to the period 1703.

Finally, when the write command is accepted, the data storage devices 4a to 4e respectively write the reconstructed data output from the controller 3, and the process is completed. Here, the period 1705 'is the data storage device 4a.
4e is the time from the start to the end of writing the data output from the controller 3.

When there is no read request from the outside, the write command from the controller 3 is accepted immediately after the end of the period 1702, and the data storage devices 4a to 4e write the data output from the controller 3. Here, the period 1705 is the time from when the data storage devices 4 a to 4 e start writing the data output from the controller 3 to when the data storage devices 4 a to 4 e finish writing, and the magnitude thereof is the period 17
Equal to 05 '.

Comparing the case where there is no read request from the outside with the case where there is a read request from the outside, the reconfiguration operation is completed in the period 170 '(the period 1703) when the read request is received from the outside. ) + Excessive time t2.

However, the reconstruction operation period 1706
Since the data reading 1703 from the outside and the data reading 1703 from the outside are performed in parallel, the time for the reconfiguration operation when there is data reading from the outside is significantly shortened as a whole.

Next, referring to FIG. 17C, the reconstruction wr
The operation when a read request is issued during execution of the ite state 53 will be described.

From the start of the operation, the data storage device 4a
Up to the reconstruction operation of the cluster data read from 4e to 4e is the same as the case shown in FIG. 17B, and in this case, the reconstruction operation is completed, and the controller 3
A write command is output to the data storage devices 4a to 4e. Here, the period 1702 is the time from the start to the end of the reconfiguration operation, and the period 1703 is the time until the write command is output and accepted by the data storage devices 4a to 4e.

When the write command from the controller 3 is accepted, the data storage devices 4a to 4e cause the controller 3 to operate.
Write the reconstructed data output from each. Here, in the period 1705, the data storage devices 4a to 4e.
Is the time from the start of writing the data output from the controller 3 to the end thereof.

Here, when a data read request 1710 is transmitted from the host device 1 to the data storage array device 2 during the period 1705, the controller 3 waits for the read request 1710 to be accepted, The reconstructed data output from the controller 3, which is currently being executed, is continuously written.

Writing of the reconstructed data is completed,
When the period 1705 ends, the controller 3 accepts the data read request 1710, and the data storage devices 4a to 4a.
4e is started to read data from the host device 1.

The data storage devices 4a to 4e cause the host device 1 to read data.

In the case of FIG. 17C, comparing the case where there is no read request from the outside with the case where there is a read request from the outside, the reconfiguration operation ends when the read request from the outside ends more than when it does not. Also, the period 1706 for reading data from the outside is longer.

As described above, 3 shown in FIGS. 17 (a) (b) (c)
As described above separately, when there is a data read request from the outside, (1) the data storage device 4
While the cluster data is being read from a to 4e to the controller 3 for the reconstruction process, the cluster data is read out and then the data is read to the outside in parallel with the reconstruction of the cluster data. Let me do it
(2) Data storage devices 4a to 4 in the controller 3
While the cluster data read from e is being reconstructed, the data is read out in parallel with the reconstructing of the cluster data, and (3) the data storage devices 4a to
While 4e is writing the reconstructed data from the controller 3, the rewriting of the reconstructed data is completed, and then the data is read out to the outside. I am trying.

In each of these operations (1), (2), and (3), the data read to the outside is performed within each transition of the reconfiguration read 52, the data reconfiguration 55, and the reconfiguration write 53. As a result, at least (1) (2)
In this operation, since the access response performance to the external access request is improved, it is possible to improve the overall external access response performance.

Next, the operation when a data write request is issued from the outside during the reconstruction process will be described.

First, if the data to be externally requested for writing is not the same as the cluster to be reconstructed, the same operation as the above-mentioned read request is performed. That is, the “read request” during execution of the reconfiguration read state 52 of FIG. 17A is changed to the “write request”, and the “read request” during execution of the data reconfiguration state 55 of FIG. 17B is executed. 17 "is replaced with" write request ", and the" read request "during execution of the reconfiguration write state 53 in FIG. 17C is replaced with" write request ".

Secondly, when the data which is the object of the write request from the outside is the same as the cluster which is the object of the reconstruction, the following is done.

(A) When a write request is issued from the host device 1 during execution of the reconfigured read state 52, the controller 3 stops reading cluster data from the data storage devices 4a to 4e, and In response to a write request from the device 1, the data from the outside is overwritten and recorded in the portion of the data storage devices 4a to 4e where the cluster data is recorded. At this time, the controller 3 completes the reading of the cluster data from the data storage devices 4a to 4e, and then overwrites and records the external data on the portion of the data storage devices 4a to 4e where the cluster data is recorded. May be. At this time, the redundancy calculation unit 34 discards the read cluster data.

(B) During execution of the data reconstruction state 55,
When there is a write request from the host device 1, the controller 3 immediately cancels the data reconstruction, and according to the write request from the host device 1, the data storage device 4a.
Data from the outside is overwritten and recorded in the portion where the cluster data of 4e is recorded.

(C) If a write request is issued from the host device 1 during execution of the reconfigure write state 53, the controller 3 stops writing the reconfigured data to the data storage devices 4a-4e. Then, in response to a write request from the host device 1, external data is overwritten and recorded in the portion of the data storage devices 4a to 4e where the cluster data is recorded. At this time, the redundancy calculation unit 34 discards the cluster data being written. The data from the host device 1 may be overwritten after the reconfigure write state 53.

In the above description, FIG. 17 (a)
The operation shown in (1) has been described as being performed within the period 1704 during which the read request or the write request 1710 from the host device 1 is reading the cluster data from the data storage devices 4 a to 4 e to the controller 3. Period 17 from when the controller 3 outputs the read command to when the data storage devices 4a to 4e accept it.
It may be performed during 01.

Next, the process of acquiring and releasing the cluster access right in the above operation will be described with reference to FIGS. 15 and 16.

FIG. 15 is a flowchart of the operation when acquiring the cluster access right.

First, the cluster control unit 32 receives an application for the acquisition of the class access right in the cluster access right acquisition routine, and this is from the host execution unit 31 or the cluster control unit 32. It is determined whether or not (S1201). Host execution unit 31
If the cluster is to be accessed, it is determined whether or not the cluster to be accessed is a “reconfiguration write”, that is, a reconfiguration processing Write is in progress (S1202). If the cluster is in the process of being reconfigured, the write count of the cluster in the cluster management table 36 is incremented by 1 (S1203), and the access state of the cluster is set to "access" (S1204). 31
Issue permissions to. On the other hand, in S1202, the access state of the cluster to be accessed is "reconfiguration w
If it is not “Write”, the access permission is issued immediately in S1205.

On the other hand, in S1201, if the application for acquiring the class access right is from the cluster control unit 32, it is determined whether or not it is in the "standby" state for access (S1206). In the case of “standby”, it is determined whether the access from the cluster control unit 32 is a read of the reconfiguration processing (S120).
7). If it is a reconfiguration read, the access state of the cluster is set to “reconfiguration read” (S120
8), issue access permission to the cluster control unit 32 (S1209).

On the other hand, in S1206, when the access state of the cluster is not "standby", it is determined that the cluster is already in the reconfiguration access state, and therefore no access permission is issued. Also, S1207
In the case where the access from the cluster control unit 32 is not the read (read) of the reconfiguration process, it is determined whether the access is the write (write) of the reconfiguration process (S1210). If it is a reconfiguration write, the access state of the cluster is set to "reconfiguration write" (S1211), and access permission is issued to the cluster control unit 32 (S1212). At this time, S121
At 0, access permission is not issued because access is not required unless it is reconfiguration Write.

Next, the operation for releasing the cluster access right will be described with reference to the flowchart in FIG.

First, in the cluster access right release routine, the cluster control unit 32 receives an application for release of the cluster access right, which is the host execution unit 31.
It is determined whether or not it is from the cluster control unit 32 (S1301). Host execution unit 3
If it is from 1, the access is "write
e ”, that is, whether it is for ending the external write processing (S1302). External writ
If it is to end the e process, the write count of the cluster in the cluster management table 36
Subtract 1 from t (S1303), and then write co
It is determined whether the remaining amount of unt is 0 (S130).
4). If the remaining write count is 0, the access state is set to "standby" (S1305), and the process ends. At this time, if the access is not "write" in S1302 or the value of write count is not 0 in S1304, the process ends.

In addition, in the operation of S1301, the application for releasing the cluster access right is issued by the cluster control unit 32.
If the access is from
e ”, that is, whether it is for ending the reconstruction write processing (S1306). Reconstruction wr
If it is for ending the item processing, the access state of the cluster is set to "standby" (S130).
7), the process ends. If it is not for ending the reconstruction write processing, the processing directly ends.

In the above operation, the steps of reading the data to be reconstructed, reconstructing the data, and writing the reconstructed data are as follows:
They are separated by interposing the steps of confirming the cluster access right, and they are processed separately in terms of time. This reconfiguration process is repeatedly executed until there are no clusters that need reconfiguration, and if there are no clusters that need reconfiguration, the state transitions to the "normal access" state. This reconfiguration processing is executed in parallel with external Read and external Write, which are external accesses to the data storage device.

External Read / External Wri described above
Regarding te and reconfiguration processing, the operation timing of the conventional reconfiguration processing and the operation timing of the reconfiguration processing in the present embodiment will be described.

FIG. 7 shows the relationship between the conventional reconfiguration processing and the data access request processing from the host device. A read operation and a write operation (wri) of the conventional reconstruction process
te) operation is continuously executed, and when an access request is input from the host device during that time, the operation is put in a wait state until the read operation and the write operation are completed, and executed after the reconfiguration processing is completed. .

On the other hand, the timing of this embodiment is rea for cluster reconstruction as shown in FIG.
By dividing and processing the d operation and the write operation, the access request from the host device can be executed during this period, and the delay is reduced. This makes it possible to respond at high speed to an access request from an external host device. Although the data size and time to access are not specified in this figure, for example, 1MB
If the cluster is managed with a size of (megabyte), it takes several tens of ms or more if Read and Write for reconstruction are combined. Even in such a case, read operation and wri for cluster reconfiguration
The delay can be halved by splitting the te operation.

Next, the operation timing for guaranteeing the correctness of the write data will be described with reference to FIGS.

In FIG. 8, Write from the outside of reconstruction is performed.
Although the operation is executed, the host device 1
If the write-requested cluster from the data storage array device 2 and the cluster reconstructed in the data storage array device 2 match, the cluster written to the data storage device by the access request from the host device 1 is the last valid one. The data. However, in the case of the operation example shown in FIG. 8, since the write operation of the divided reconfiguration processing is performed later, the data obtained by the reconfiguration processing overwrites the data written from the host device 1. And the legitimacy is not guaranteed.

The present embodiment is for avoiding such a situation as described above. The acquisition / release of the cluster access right shown in FIGS. 15 and 16 and the access for each cluster shown in FIG. 6 are carried out. By managing the state, writing (Write) to the same cluster is inspected, and when it is the same cluster, writing (Write) of reconfiguration is canceled.

The operation timing is shown in FIG. The access right check is executed next to the read of the reconfiguration, and when it is determined that the write is the write to the same cluster, the write of the reconfiguration is canceled (Write).
Cancel). This ensures the validity of the data.

As described above, the online reconfiguration processing apparatus and the online reconfiguration processing method in the data storage array apparatus according to this embodiment can guarantee normal data even in the event of a failure due to the redundant configuration. It is possible to speed up the response time to the host device even during the reconfiguration process after replacing the data storage device.
At this time, the correctness of the data writing from the host device can always be guaranteed by canceling the data writing in the reconstruction process.

That is, in the present embodiment, since the state of each cluster is managed by the cluster management table 36, it is possible to take the state according to the external access and the reconfiguration access. When there is a request for external access to the external access, the external access can be interrupted for the operations of reading data for reconfiguration, reconfiguring data and writing reconfigured data.

Since the consistency of data is assured by the management of the access right of the cluster, the external access and the access for reconstruction may be multi-divided by the normal multi-task processing. In practice, the multitasking OS processes the external access and the reconfigured access in a time-sharing manner.

In the above description, the cluster management table 36 is described as being held in the cluster controller 32, but it may be held by the array controller 33. Further, although the cluster access right management illustrated in FIGS. 15 and 16 is described as being performed by the cluster control unit 32, the array control unit 3 is described.
3 may manage it.

(Second Embodiment) Next, referring to FIGS. 10 and 11, as an online reconfiguration processing method and an online reconfiguration processing apparatus according to a second embodiment of the present invention, delay processing in division of reconfiguration steps is performed. Will be described.
Since the configuration of the online reconfiguration processing apparatus according to this embodiment is the same as that of the first embodiment, the description of FIG. Further, FIG. 10 shows a timing chart when the delay process of the reconstruction step is not performed. Ste
p1 is a reconstruction read, Step2 is a data reconstruction process, and Step3 is a reconstruction Write. In order to increase the average bandwidth of access from the host device 1 to the data storage array device 2, it is necessary to limit the bandwidth of reconstruction. As shown in FIG. 11, Step 1, Step 2, and St
The bandwidth for reconstruction is limited by setting transition times of T1 and T2 between ep2 and Step3, respectively. The timing at which the cluster controller 32 issues an access request to the array controller 33 is adjusted by T1 and T2.

The transition times T1 and T2 are determined by the external host device 1
It changes according to the frequency of data access. For example, when the data access frequency is high, it is determined that the bandwidth required by the external host device 1 is large, and T1 and T2
Lengthen the time. Further, instead of the frequency of occurrence of data access, the average bandwidth of data access from the external host device 1 is used, and when the average bandwidth is large, T1 and T
Make 2 longer. If no data access from the external host device 1 continues, T1 and T2 are reduced.
At this time, the values of T1 and T2 may be zero. As a result, the bandwidth can be guaranteed for external access, and when there is no external access, efficient reconstruction can be performed by eliminating the delay between each step of the reconstruction processing. Although the time between Steps 1 to 3 is set as T1 and T2, the time from the identification of the cluster to be reconfigured to the start of the processing of Step 1 is T0.
Similarly, for 0, the transition time can be set according to the external access band to control the reconfiguration band. The transition time may be adjusted to the same length during all the steps, or may be adjusted independently for each step. Also, when there is a data access during the execution of a certain step, the transition time may be set between that step and the next step, or the transition time between that step and the next step. You may set in between.

As described above, according to the present embodiment, it is possible to limit the bandwidth of reconfiguration access according to the bandwidth of access from the outside, and guarantee the bandwidth of the access from the outside. In addition, it is possible to efficiently perform reconfiguration while there is little external access.

In the access right check to the data storage device, if there is an access request from the host device 1 and a reconfiguration access request from the cluster controller 32 at the same time, the access request from the host device 1 is always prioritized. By doing so, it is possible to guarantee the responsiveness to the host device 1.

In each of the above embodiments, the controller 3 of the data array storage device 2 corresponds to the arbitration device of the present invention. Further, the host unit 31, the array control unit 33, and the data storage device management unit 35 are the components of the present invention.
It corresponds to the process determination means and the arbitration means .

Further, the present invention is a program for causing a computer to execute all or some of the steps (or steps, operations, actions, etc.) of the above-described online reconstruction processing method of the present invention, It is a program that operates in cooperation with a computer.

The present invention is a medium carrying a program for causing a computer to execute all or some of the operations of all or some of the steps of the above-described online reconstruction processing method of the present invention, which is readable by the computer. The read program is a medium that executes the operation in cooperation with the computer.

Note that some of the steps of the present invention (or
(Process, action, action, etc.) means some of the plurality of steps, or means a part of the action of one step.

[0124] Further, some devices (or elements, circuits, parts, etc.) of the present invention mean some of those devices, or one device. , Means a part of means (or an element, a circuit, a part, etc.) or means a part of the function of one means.

Further, a computer-readable recording medium in which the program of the present invention is recorded is also included in the present invention.

Further, one usage form of the program of the present invention may be a mode in which the program is recorded in a computer-readable recording medium and operates in cooperation with the computer.

Further, one usage form of the program of the present invention may be a mode in which it is transmitted through a transmission medium, read by a computer, and operates in cooperation with the computer.

Further, the recording medium includes a ROM and the like, and the transmission medium includes a transmission medium such as the Internet,
Light, radio waves, sound waves, etc. are included.

The computer of the present invention described above is
The hardware is not limited to pure hardware such as a CPU, and may include firmware, an OS, and peripheral devices.

As described above, the configuration of the present invention may be realized by software or hardware.

[0131]

As is apparent from the above description, according to the data storage array device of the present invention, the reconstruction process is divided and the access from the external host device is accepted during the division process. High-speed response is possible. When an external access process (Write) actually occurs during Read / Write of the reconfiguration process, and
In case of access to the same cluster, Wr of reconfiguration processing
It is possible to guarantee the order of data by canceling ite. Further, it is possible to control the bandwidth of the reconfiguration by providing a delay between the read of the reconfiguration processing, the restoration and the processing of the write, and to guarantee the bandwidth of the access from the external host device.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a data storage array device according to a first embodiment of the present invention.

FIG. 2 is a block diagram showing a management area of the data storage array device according to the first embodiment of the present invention.

FIG. 3 is a diagram for explaining processing contents of a controller 3.

FIG. 4 is a diagram showing an example of a configuration of a cluster management table 36.

FIG. 5 is a state transition diagram of each cluster.

FIG. 6 is a state transition diagram of access states of each cluster.

FIG. 7 is a conventional reconfiguration access timing chart.

FIG. 8 is a reconfigurable access timing chart according to the second embodiment of the present invention.

FIG. 9 is a reconfiguring access timing chart in the case of the same cluster access according to the second embodiment of the present invention.

FIG. 10 is a timing diagram of the reconstruction processing steps.

FIG. 11 is a timing diagram of a reconstruction process step with delay.

FIG. 12: Reading to the data storage device 4 (Read)
It is a figure which shows an access flowchart.

FIG. 13 is a diagram illustrating writing to the data storage device 4 (Writ
e) It is a figure which shows an access flowchart.

FIG. 14 is a diagram showing a reconfiguration access flowchart of the cluster control unit 32.

FIG. 15 is a diagram showing a flowchart showing an operation of acquiring a class access right.

FIG. 16 is a diagram showing a flowchart showing an operation of releasing a class access right.

FIG. 17 (a) is a diagram showing a timing chart for explaining the operation of the first embodiment of the present invention. FIG. 6B is a diagram showing a timing chart for explaining the operation of the first embodiment of the present invention. (C) Embodiment 1 of the present invention
5 is a diagram showing a timing chart for explaining the operation of FIG.

[Explanation of symbols]

1 Host device 2 Data storage array device 3 controller 4 Data storage device 31 Host Execution Unit 32 cluster controller 33 Array controller 34 Redundancy calculator 35 Data Storage Device Management Unit

─────────────────────────────────────────────────── ─── Continuation of the front page (56) Reference JP-A-8-221217 (JP, A) JP-A-11-85576 (JP, A) JP-A-11-119921 (JP, A) JP-A-11- 265308 (JP, A) JP 2000-56935 (JP, A) JP 2000-276305 (JP, A) (58) Fields investigated (Int.Cl. ⁷ , DB name) G06F 3/06-3/08 G06F 12/16

Claims (4)

(57) [Claims] 1. A plurality of segments formed by dividing a recording area of a plurality of data storage devices, and a plurality of parity groups obtained by grouping the segments of each of the plurality of data storage devices using a predetermined algorithm. In the data storage array device storing redundant data in at least one of the segments in the plurality of parity groups, the reconfiguration process is performed when an access from an external host device occurs during the reconfiguration process. An arbitration method for performing arbitration with a necessary access, wherein a read access request from an external host device during the reconfiguration processing or a write access request to a parity group different from the parity group under the reconfiguration processing is made. And a process determination step for determining a process in which the reconstruction process is in progress; In step, the reconstruction process is read process, or when it is determined that the reconfiguration process, a read access or a write access from the external host device, by interrupting before the writing process in the reconstruction process , By the arbitration step,
In parallel with the reconstruction process in the reconstruction process, the external
Characterized by enabling access from the host device
Arbitration how. 2. A plurality of segments formed by dividing a recording area of a plurality of data storage devices, and a plurality of parity groups obtained by grouping the segments of each of the plurality of data storage devices using a predetermined algorithm. And is required for the reconfiguration process when an access from an external host device occurs during the reconfiguration process of the data storage array device that stores redundant data in at least one of the segments in the plurality of parity groups. An arbitration device that performs arbitration with different access, when a read access request from the external host device during the reconfiguration processing or a write access request to a parity group different from the parity group being reconfigured A process determining means for determining a process in which the reconstruction process is in progress, and the process determining means. When it is determined that the reconfiguration process is in the reading process or the reconfiguration process, the arbitration unit interrupts the read access or the write access from the external host device before the writing process in the reconfiguration process. And a reconstructing process in the reconstructing process by the arbitration means.
Access from the external host device at the same time
An arbitration device. 3. A read access request from an external host device during the reconfiguration process of the arbitration method according to claim 1.
Or when there is a write access request to a parity group different from the parity group being reconfigured,
From the external host device, when the reconfiguration process is in the reading process or the reconfiguration process in the process determination step of determining the process in which the reconfiguration process is in progress, and the process determination step. A program for causing a computer to execute an arbitration step of interrupting read access or write access before the writing step in the reconfiguration processing. 4. A recording medium carrying the program according to claim 3 , which can be processed by a computer.