JPH08286843A - Method for decoding data in disk array - Google Patents

Abstract

PURPOSE: To prevent deterioration in the speed of a processing unit even on the occurrence of a fault in a disk device by conducting prescribed error decoding processing depending on a degree of access requests of file data distributed in plural disk devices and transferring the decoded data. CONSTITUTION: A data transfer control device 7 divides data into the units of blocks to distribute a block of data to plural disk devices 11 at data write and generates a parity, and integrates the distributed data at read. Furthermore, an access request monitor device 5 discriminates quantity of access requests given to a disk array controller 2, and when part of the plural disk devices 11 has a fault, a data decoding control device 6 decodes data based on the data and the parity of the other normal disk devices. When the quantity of the access requests is not large, the error decoding is applied to the data and the resulting data are transferred, and when the quantity of the access requests is large, the corresponding data are read from the disk device and the resulting data are transferred as they are.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data restoration processing method in a disk array system of a data distributed recording system when a disk device for storing data requested in response to a data read request has a failure. is there.

[0002]

2. Description of the Related Art Due to recent advances in semiconductor technology, the performance of a central processing unit has been dramatically improved, and, in keeping with this, there has been a demand for higher speeds in peripheral external storage devices. By the way, there is a limit to the speeding up of a magnetic disk device that involves mechanical operation as an external storage device. As a means for improving this, a disk array system which accesses a plurality of magnetic disk devices in parallel has recently attracted attention. A disk array stores data in blocks of a certain amount in a distributed manner in multiple disk devices (data striping), and accesses these multiple disk devices in parallel to read / write multiple divided data simultaneously. However, the access speed is being increased. This method is particularly effective for accessing large amounts of data such as moving images and audio.

Further, in a disk array called RAID5, when data is distributed and stored, parity is generated and similarly stored in the disk device, so that a failure of the disk device or a failure such as a defective sector occurs. Then, it deals with the case where the data cannot be read. That is, it is possible to restore the data stored in the failed disk device by parity calculation from the data and the parity distributed in other disk devices. FIG. 4 is a diagram showing a flow of processing of a read request in the conventional RAID 5 disk array. In the RAID5 disk array, when there is no failure in the disk device, the data is read from a plurality of disk devices that store the requested data, and the data transfer control mechanism reconstructs the data into a group of data and transfers the data to the host device. Forward. If there is a failure in some of the multiple disk devices that store the requested data, the failure will occur based on the normal disk device data and parity, regardless of the access request at that time. The data stored in the disk device is restored, and the data is transferred to the higher-level device while being reconstructed into a set of data together with other data.

FIG. 3 is a diagram showing how data is stored in a disk device constituting a RAID 5 disk array. When writing data of file A, the disk array controller divides file A into blocks of a certain amount, and in this example, {a1, a2, a
3, a4, a5, a6} and divided into six blocks, and stored over a plurality of disk devices as shown in the figure. At that time, parities such as P1 and P2 are also generated and stored. At the time of a read request for the file A, the disk array control device causes each block {a1,
A read command is issued to the disk device that stores a2, a3, a4, a5, a6}, and the data is transferred to the higher-level device while being reconfigured as a group of files A by the data transfer control mechanism in the disk array control device. .

When a read request for the file A is made, the file A
A plurality of disk devices for storing disk
2, a part of Disk3, Disk4, Disk5, for example, Disk3 has failed, blocks of file A stored in a normal disk device {a1, a2, a
4, a5, a6} is read out, and P1 which is the parity is also read out, and the exclusive OR of {a1, a2, a4, P1} is calculated in the disk array controller to restore the data of a3. Then, a3 is restored to the position of x in the file A on the Read side of FIG. 3, and is reconfigured as a group of files A and transferred to the host device.

In order to clarify the position of the present application, two processes for calculating the data stored in the disk device having the failure from the data thus stored in the normal disk device and the parity are described. I will roughly divide it. That is, after replacement of the failed disk device or in the case of the spare disk device, the case where the data of the entire failed disk device is calculated and returned to the original is called recovery processing. In addition, when recovery processing has not been performed at the time of data read request from the host device, only the data stored in the failed disk device among the requested data is calculated and the correct data is given. Then it is called restoration processing.
The present invention relates to the latter restoration process for providing correct data for the time being. Regarding the former recovery process,
A method has been proposed in which control is performed so that recovery processing is performed when there is no access request at all, or when there are few requests, and JP-A-5-165581 is one example.

[0007]

In the conventional disk array, when a disk device fails, data is always restored and transferred regardless of the amount of access requests held at that time. When there are many access requests, reading data and parity from other disk devices and restoring the data of the failed disk device by calculation hinders processing for other accesses, increasing the processing speed of the disk array system. There is a problem of lowering it. In a disk array that handles video and audio data, it is important to continuously transfer continuous data at a constant speed, but when there are many access requests, data restoration processing is performed due to some disk device failures. Doing so will affect not only the data being restored but also other unrelated data.
On the other hand, with regard to data such as moving images and voice, even if some data is lost, the image quality and sound quality are temporarily degraded, but the meaning of the data as a whole is not lost.

The present invention has been made to solve the above problems, and an object of the present invention is to obtain a data restoration control method that suppresses a decrease in processing speed even when a disk device has a failure.

[0009]

In the disk array data restoration method according to the present invention, there is a read request for file data distributed and written in a plurality of disk devices, and at the time of this read request, a failure is detected in the disk device. Then, the step of checking the degree of access request within the set time, the step of transferring the data by performing a predetermined error restoration process when the degree of access request is small, and the step of directly transmitting from the disk device when the degree of access request is large. Reading and transferring the corresponding data.

In addition to the basic method, when a data importance flag is further provided in the read request and a read request in which the data importance flag is set is received,
Even if there are many access requests, a step of performing predetermined error recovery processing and transferring data is added.

Furthermore, when there is an error and the data is transferred as it is without the error restoration process, the lost data is not replenished and transferred.

Furthermore, when there is an error and the data is transferred as it is without performing the error restoration processing, the lost portion is replaced with the predetermined data and transferred.

[0013]

According to the disk array data restoration method of the present invention, even if there is a failure in the disk device, data transfer is performed without error restoration processing if the access request degree is high, and error restoration is performed if the access request degree is low. Process and transfer.

Furthermore, if there is an error and the importance flag is set, error recovery processing is performed and the data is transferred.

Furthermore, when there is an error and the data is transferred, the data is transferred while the error portion is lost.

Furthermore, when there is an error and the data is transferred, the error portion is replaced with predetermined data and the data is transferred.

[0017]

【Example】

Example 1. FIG. 1 is a diagram showing the configuration of an embodiment of the present invention. The disk array control device 2 in the figure is a host device I /
It has an F control mechanism 3, an access request processing mechanism 4, an access request monitoring mechanism 5, a data restoration control mechanism 6, a data transfer control mechanism 7, and a lower device I / F control mechanism 8. Host device I
The / F control mechanism 3 performs control for exchanging information such as an access request command and data from the higher-level device 1. The access request processing mechanism 4 analyzes the access request from the higher-level device, and gives an instruction to the lower-level device I / F control mechanism 8, the data transfer control mechanism 7, the data restoration control mechanism 6, etc., if necessary. At the time of writing data, the data transfer control mechanism 7 divides a set of data into a plurality of disk devices 11, divides the data into blocks of a certain amount, and further generates parity. Further, at the time of reading data, the data divided into a certain amount of blocks and distributed to a plurality of disk devices are integrated into one data.

The access request monitoring mechanism 5 is a mechanism for judging the number of access requests coming to the disk array control device 2. For example, in the case of a system compatible with command queuing, the number of processes accumulated in the queue is determined. Originally monitor the amount of access requests. Also, in the case of a system that does not support command queuing and sequentially executes each request, the number of access requests that came per unit time is counted to determine the amount of access requests. Monitor. When there is a failure in a part of the plurality of disk devices 11 storing the data requested to be accessed, the data restoration control mechanism 6 uses the data and parity stored in other normal disk devices as a basis. It has a mechanism to restore the data stored in the failed disk device by calculation. Further, when there are many access requests, instead of not performing restoration, processing such as embedding other data is performed.

The lower I / F control mechanism 8 generates an instruction to a plurality of disk devices 11 for storing the requested data in a distributed manner in response to the access request from the upper device analyzed by the access request processing mechanism 4. , Issues an instruction to each disk control mechanism 9. In the example of FIG. 1, the lower I / F control mechanism 8 includes five disk control mechanisms 9, and each disk control mechanism 9 is connected to a plurality of disk devices 11 by a disk interface bus 10 such as SCSI, and these plurality of disks are connected. The device 11 constitutes a disk array 12.

FIG. 2 is a diagram showing a processing flow showing an embodiment of the data restoration method of the present invention. First, processing of a read request when there is no failure in the disk device will be described. The access request generated by the host device is the host device I / F
The access request processing mechanism is reached via the control mechanism. In step S1, the access request processing mechanism checks in which disk device the requested data is stored, and determines whether a failure has occurred in those disk devices. If all the disk devices storing the requested data are normal, the data transfer control instruction is given to the lower-level device I / F control mechanism in step S2, and the data is distributed and stored in the plurality of disk devices. The data is read in parallel, and the data divided by the data transfer control mechanism is transferred to the host device again via the host device I / F control mechanism while connecting the data divided in step S3 in the original order.

Next, the read request processing when the disk device has a failure and the access request amount is small will be described. Until the access request reaches the access request processing mechanism, the processing is the same as the normal processing. The access request processing mechanism checks the disk devices storing the requested data, and if a failure occurs in those disk devices in step S1, the access request monitoring mechanism is inquired about the current access request amount. . The access request monitoring mechanism monitors the amount of access requests based on the number of queued access requests, the number of access requests processed per unit time, or the like, and step S11
Then, the access request amount is determined depending on whether this value is larger or smaller than a certain value. If it is determined that the access request amount is small, the access request processing mechanism instructs the lower I / F control mechanism to read the data from the normal disk device in step S12. At the same time, in order to restore the data stored in the failed disk device, the corresponding parity data is also read in step S12. The data restoration control mechanism restores the data stored in the failed disk device in step S13 by calculation based on the read data and parity from the normal disk device, and in step S14, the data transfer mechanism. Send to. In this way, the data read from the normal disk device and the data restored by the data restoration control mechanism are collected in the data transfer control mechanism, the divided data are connected in the original order, and the host device I / F control mechanism is connected. It is transferred to the higher-level device via. In the case of the file A in FIG. 3 as an example, if the Disk3 fails and the data cannot be read at all, a3 is restored at the position of x in the drawing based on other data and parity and transferred to the host device. Has been done.

Next, processing of a read request when there is a failure in the disk device and the amount of access requests is large will be described. The process up to the point of determining the presence / absence of a failure in the disk device and inquiring the access request amount is the same as in the case where the access request is small. If it is determined in step S11 that the access request amount is large, the access request processing mechanism sets the importance flag, that is, the restoration flag. Here, when the series of data (file data) requested to be read is considered to be important and the restoration flag is ON, the data cannot be partially lost. Therefore, it is determined that the restoration process needs to be performed regardless of the amount of access requests, and thereafter, the same process as when the access requests are small is performed. Restore flag is OFF,
In other words, in response to a read request of which the importance of the file data is not so high, the lower device I / F control mechanism 8 is instructed to read the data in step S22, and the necessary data is obtained in step S23. In this case, for example, the data is sent without any data and is left blank. In step S24, the data restoration control mechanism 6 reconstructs these data into a set of data (file data), and the data transfer control mechanism 7
Is sent to the higher-level device 1, for example. The importance flag (restoration flag) is not always necessary and may not be used. In that case, the Yes route in step S11 immediately leads to step S22, and the disk device reads data. Further, in this example, the restoration flag is determined after the access request amount is determined, but it may be determined before the access request amount is determined.

Embodiment 2 FIG. Another special process when the disk device has a failure, the access request amount at that time is large, and the restore flag is present, the flag is OFF, or the restore flag is not present will be described. In this case, the data restoration control mechanism performs another process without restoring the data stored in the failed disk device. That is, various processes are performed in step S22 of FIG. First, in the first example,
The data stored in the faulty disk device is removed, only the data from the normal disk device is connected, and the data is transferred to the host device via the host device I / F control mechanism by the data transfer control mechanism. It In FIG.
If Disk3 fails, the block corresponding to x in the figure is not transferred to the upper device, and {a1, a2, a4,
Only 5 blocks of a5, a6} are transferred. When a defective sector exists in the disk device, only the data in the defective sector is extracted and transferred.

In the second example, the faulty data is removed, the data from the normal disk device is connected, and appropriate data such as "0" data is added by the amount of data that is lost at the end. To do. This data is transferred to the host device via the host device I / F control mechanism by the data transfer control mechanism. In FIG. 3, assuming that the Disk3 has failed, the block corresponding to x in the figure is filled with the next block a4, and the blocks are sequentially packed in order, and appropriate data such as "0" data is stored in the block next to a6. With the addition of
Transfer according to the number of bytes transferred from the host device. If there is a defective sector in the disk device, the data is removed, and appropriate data such as "0" data is added for the amount of data that was removed last and transferred.

In the third example, appropriate data such as "0" data is embedded and transferred at the position of the data having the failure. The data thus generated is transferred by the data transfer control mechanism to the host device via the host device I / F control mechanism. In FIG. 3, if the Disk 3 fails, appropriate data such as “0” data is embedded and transferred in the block corresponding to x in the drawing. If a bad sector exists in the disk device, "0" is set at that position.
Transfer is performed by embedding appropriate data such as data.

Further, in the fourth example, the same data as the data read from the previous disk device is embedded in the position of the faulty data and transferred. That is, in FIG. 3, if the Disk3 fails, the same data as the previous a2 is embedded and transferred to the block corresponding to x in the drawing.

In the fifth example, the same data as the data read from the next disk device is embedded and transferred at the position of the defective data. That is, in FIG. 3, assuming that the Disk3 has failed, the same data as the next a4 is embedded and transferred to the block corresponding to x in the drawing.

In the sixth example, the same data as the previous sector is embedded and transferred at the position of the data written in the defective defective sector. Also, in the seventh example, at the position of the data written in the defective defective sector,
The same data as the next sector is embedded and transferred.

[0029]

As described above, according to the present invention, the restoration process is changed depending on the amount of access requests. Therefore, even if there is a failure in the data requested to be accessed, the amount of other access requests is changed. If the number is small, the restored complete data can be transferred. On the other hand, if the amount of other access requests is large, the data is transferred without performing the restoration process, so that the access speed is increased.

Further, if the importance flag (restore flag) is set, the restoration process is performed even if the amount of access requests is large, so that the quality of important data can be prevented from being deteriorated.

Furthermore, since no extra data is generated and data is transferred as it is, there is no delay in access,
The upper device has an effect that error detection is easy and arbitrary data can be supplemented.

Furthermore, since the predetermined data is added and transferred, there is an effect that the erasure process suitable for the system can be selected.

[Brief description of drawings]

FIG. 1 is a diagram showing an example of a configuration of a disk array system that realizes a data restoration method of the present invention.

FIG. 2 is a diagram showing a processing flow of a data restoration method of the present invention.

FIG. 3 is a diagram showing a state of data and file data stored in a disk device forming a disk array.

FIG. 4 is a diagram showing a conventional data restoration processing flow.

[Explanation of symbols]

1 host device, 2 disk array control device, 3 host device I / F control mechanism, 4 access request processing mechanism, 5
Access request monitoring mechanism, 6 data restoration control mechanism, 7
Data transfer control mechanism, 8 Lower device I / F control mechanism, 9
Disk control mechanism, 10 disk interface bus, 11 disk device.

Claims (4)

[Claims] 1. A step of checking a degree of an access request within a set time when there is a read request for file data distributed and written to a plurality of disk devices and a failure is detected in the disk device at the time of the read request. If the access request degree is low, the disk is provided with a step of performing a predetermined error recovery process to transfer the data, and if the access request degree is high, the step of directly reading the corresponding data from the disk device and transferring the disk. Array data recovery method. 2. Further, a data importance flag is further provided in the read request, and when a read request in which the data importance flag is set is received, predetermined error restoration processing is performed even if there are many access requests. The method of restoring data in a disk array according to claim 1, further comprising a step of transferring data by means of data transfer. 3. The data restoration of the disk array according to claim 1, further comprising: when the data is directly transferred without any error restoration processing even if there is an error, the lost data is not replenished. Method. 4. The disk according to claim 1, further comprising the step of replacing the lost data with predetermined data and transferring the data when the data is transferred as it is without any error restoration processing even if there is an error. Array data recovery method.