JP2003036146A - Disk array control system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enables a disk control part with low error detecting capability to detect such trouble that old data are read out by mistake and also detect even an error of data by using an added check code. SOLUTION: A disk array control system is characterized by that a disk array device constituted by using a plurality of disk units writes added check codes 220, 221, and 222 generated according to data written to data disks constituting a parity group to the data disks, writes all the added check codes 220, 221, and 222 to parity disks constituting the parity group, and reads added check codes of one of the plurality of data disks and of parity disks to check an error.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a failure detecting method for a disk array device, and more particularly to a technique for constructing a highly reliable disk array device by combining disk devices which are inexpensive and have a low failure detecting ability.

[0002]

2. Description of the Related Art Generally, a disk uses a magnetic recording medium or an optical recording medium, so that a part of the disk cannot be read or written. This is called a sector failure. The cause of the sector failure may be scratches on the disk or deterioration of the magnetic material. In the disk device, ECC (E
By adding a error correcting code, it is possible to recover from a certain degree of sector failure. Also, depending on the type of ECC code to be added, the number of bits in a sector in which an error can be corrected changes.

In a low-priced drive, since the above-mentioned sector failure is relatively likely to occur, a high speed control is realized by operating a plurality of disks in parallel, and redundant data called parity is called a parity disk. By storing on a specific disk, data is stored by any chance 1
Even if one disk fails, a disk control method called RAID that can reproduce the data of the failed disk from other disks and the parity of the parity disk and can improve the reliability of disk failure is proposed. Was done.

Regarding this RAID, "A Case
for Redundant Arrays of
Inexpensive Disks (RAI
D) "; In Proc. ACM SIGMOD, Ju
Learn more about ne 1988 (published by the University of California, Berkeley). RAID has levels 1 to 5 depending on the method of storing the parity, but the level 1 is currently widely used in disk arrays having fault tolerance.
And level 5.

In a disk array device having the above-mentioned RAID 5 structure having a redundant structure, when a sector failure occurs, the data is restored by the remaining disk devices belonging to the same parity group, and the failure data is written to the alternate area. Had been recovered.

Here, taking the data arrangement on the four disks shown in FIG. 3 as an example, the RAID 5 configuration will be described. The data transferred to the disk array control device is, for example, 0, 1, 2. The data is divided into a fixed length of 512 bytes and transferred to the disks 1, 2 and 3, and the exclusive OR of the data 0, 1 and 2 is taken as the parity part P to form a fixed length of 512 bytes and transferred to the disk 4. To do. The data 0, 1, and 2 and the parity P form the same parity group. As shown by the arrow in FIG. 3, the data allocation to the disks 1 to 4 following the data 0, 1, 2, parity P, the data 0, 1, 2, parity P, ... Is the standard of RAID 5.

When it is attempted to read data from one disk having the RAID5 configuration, it is possible to read the data from all the data disks and the parity disks which form the parity group to detect the error. Since data is read from a plurality of disks, a large performance drop will occur.

Therefore, in the case of such a RAID device,
It is usual to add a code for error detection to predetermined data, write both of them as data, and perform reading so that the error can be detected by reading a single disk. That is, the data 0, 1, 2
In addition to the 512-byte raw data described above, a logical address LA of the disc and a check bit CH are added. Then, error detection is performed on a single disk by this LA or CH (refer to FIG. 2, the data 20 of FIG. 2 corresponds to raw data, the LA 21 of FIG. 2 corresponds to a logical address, and the RCC 23 checks. It corresponds to a bit).

Further, Japanese Laid-Open Patent Publication No. 10-171608 discloses that address information and information indicating a temporal element are added to the data part. By adopting such a data format, even if the old data in the cache register of the magnetic disk device is erroneously read, the time stamp is additionally stored in the data when converting to the data for the magnetic disk device. A data abnormality can be detected when reading or writing data.

[0010]

The technique described as the prior art is mainly aimed at detecting an error represented by a sector failure at the time of writing and reading to and from a disk medium, and detecting and recovering an error related to an interface. Is a means to do.

On the other hand, in recent years, the price of disk drives has been remarkably reduced mainly for personal computers, but
Along with that, since the cost of price reduction was emphasized, the check of the internal control system, which occurs less frequently than the failure of the medium (recording medium), was omitted, so the error detection capability of the disk controller was relatively low. Has become. Even in this case, if the error generated inside the disk is written as a simple data error such as a garbled bit of the data system, it can be detected by the error detection code of the RAID data described in the prior art. When a failure occurs in the buffer in the disk that holds the copy data, a malfunction may occur in which the data to be written is written in the wrong position.

In this kind of error, correct data is written for both the written data (data written in the wrong position) and the data left unwritten (the original data that should be overwritten and erased). If you have
In particular, the old data that was left unwritten because the data that should be written was written in the wrong location,
There is no means in the prior art to detect that this is wrong, as it is not inconsistent by itself.

[0013] According to the conventional method, a data portion including 512B of raw data, LA of position information, CH of check bits, and parity data generated from the raw data and a parity portion of LA and CH are used. RAID
, The parity part (P) and all data parts (0, 1,
Although an error can be definitely detected by performing an error check by reading with 2), considering the above-mentioned read performance degradation, if there is no problem in checking the data and check code CH in the reading of the data section, it is sent to the host as positive data. May be handled. In such a case, when the data is written in the wrong position, the disk array controller recognizes that new data has been written in the position where the data should be written. Therefore, the old data that has not been erased by overwriting and remains, will be read. Since the old data itself is the same as the CH, the error detection does not occur in the reading of the old data (since LA has no problem).

An object of the present invention is to hold an additional check code based on common data in both the data section and the parity section, and detect the additional check code in two disks, the data section and the parity section. As a result, it is possible to detect a fault in which old data is erroneously read and also to detect a data error by the additional check code.

[0015]

SUMMARY OF THE INVENTION The present invention mainly has a function or action for realizing error detection, which cannot be achieved by the prior art, by adopting the following disk array control system. (1) Identification information (CC0, CC1 or CC2 shown in FIG. 2 which is generated based on the data 20 and which is written in the data portion forming the RAID as the additional information of the parity in addition to the parity data of the parity portion. ) Is added as additional check data and writing is performed. (2) Data is always read out in units of two disks, the data section and the parity section, and the identification information of the data section and the parity accompanying information of the parity section are compared with each other to detect a fault in which old data is erroneously read. To detect.

According to the present invention, since two disks are always accessed for reading, the reading performance is somewhat lowered, but a failure to read old data by mistake can be detected.

In order to achieve the above functions and actions, the present invention mainly employs the following configurations. In a disk array device configured using a plurality of disk devices, each additional check code generated based on each data written in each data disk is written in each of the plurality of data disks forming the parity group, and A disk array control method in which all of the additional check codes are written to a parity disk that constitutes a parity group, and one of the plurality of data disks and the additional check code of the parity disk are read to perform an error check.

[0018]

BEST MODE FOR CARRYING OUT THE INVENTION A disk array control system according to an embodiment of the present invention will be described in detail below with reference to FIGS. 1, 2 and 3. FIG. 1 is a diagram showing a schematic configuration of a disk array device according to an embodiment of the present invention, FIG. 2 is a diagram showing a data format to be written to a disk according to this embodiment, and FIG. 3 is an inside disk of this embodiment. It is a figure which shows the data arrangement of.

In FIG. 1, reference numeral 110 indicates a communication path to the server 10 which serves as a host. For this communication path 110, generally SCSI or fiber channel is used. Figure 1
Reference numeral 2 denotes a disk array control device that performs control in the present embodiment, and includes a host communication control unit 121, a processor unit 122 that performs device control, a cache unit 123 that caches data, and a disk control unit 124 that controls disks. ,,. A disk 13 is connected to the disk array controller 12.

In the present embodiment, referring to FIG. 3, one piece of parity data for three pieces of data is prepared, and a RAID5 configuration in which a parity group is formed by four disks is adopted. The number is required, and Fig. 3
For simplicity, 4 units are shown. Since the data and the parity itself are distributed and arranged as shown in FIG. 3, they are not divided into a data disk and a dedicated parity disk.

FIG. 2 shows a format of data to be written in the RAID device. The data part and the parity part use the same format. Reference numeral 20 in FIG. 2 is data held by the disc, 21 in FIG. 2 is position information of the data in the disc, 22 in FIG. 2 is additional check code data which is a feature of the present invention, and 23 in FIG. 2 is data and position information. , A check code generated from the additional check code.

In this embodiment, the additional check code to be written to the parity disk is the same as the additional check code used for writing the data section (however, it is generated from the data and the position information), and the additional check code is used in the data section. The additional check code used is the additional check code used in the parity part. The check code RCC accompanying the data to be actually written is a check code for the entire data generated from the data, the position information, and the additional check code. Here, the additional check code is characterized by being a check code generated based on the data 20 as described above, and CRC (Cyclic).
A Redundancy Check) code may be used. Since each is a check code based on the data 20, it is possible to check the read data for an error by using this additional check code. Furthermore, existing data and / or position information LA is used to generate the additional check code. You are using it and do not need or use any other special sources of information. Further, an identifier that can uniquely identify writing when writing data may be used as the additional check bit.

FIG. 3 shows the data arrangement in the disc in this embodiment. Each disk 30, 31, 32, 33 forming the parity group has three data 34, 3
5, 36 and parity 37 are stored as shown,
As shown in the figure, another disk is used in a fine unit of about 64 KBytes, whereby the parity part, which increases the load, is distributed to all the disks to make the load uniform and improve the performance.

The actual operation of this embodiment will be described in detail below. In this embodiment, the communication path 110 is transmitted from the host 10 side.
When data writing is designated via the data, the data to be written sent from the host is divided into 512 bytes and buffered in the cache memory unit 123 first. Next, the control processor 122 generates data to be written on the disk 13 according to the format shown in FIG. Here, the data 20 is the data itself sent from the host, and the position information 21 is the position in the disk array of the data to be written managed internally. This consists of 8 bytes, the upper 2 bytes are the serial number (connection position information) of the disk connected to the disk array control device 12, and the lower 6 bytes indicate the write start sector number of the data in the disk.

In this embodiment, at the time of writing, the data having the additional check code 22 shown in FIG. 2 of the same format is written to both the disk where the target data is arranged and the parity disk. Since parity data is written as a check code even in the case of normal RAID, the present embodiment differs from the conventional format in that the additional check code 22 is written as ancillary data.

At this time, in the additional check code data 22 of the data to be written, there are three additional check code fields 220, which correspond to the RAID configuration of this embodiment.
221 and 222 exist, but the additional check code of the data part is only the additional check code of the position of the own data updated.
As for the additional check code of the parity data, only the additional check code corresponding to the written data is updated, and the other parts remain unchanged.

For example, when data is written in the first data area 34 of the parity group (see FIG. 3), the parity data is recalculated from the data and the additional check code of the parity data is stored in the field 2 of the data area 34.
The check code generated from the data is written in the position 220 in the field 220 of the parity data 37. Finally, this whole data (data 20, LA21,
A check code 23 having a predetermined detection capability is generated for the additional check code 22) and added as data. In the present embodiment, a check code of 8 Bytes is attached, and 552 Bytes are actually written to the disk for 512 Bytes of data. This 552 Bytes data is written from the disk controller 12 to the actual disk 13.

The disk data format described above will be described again with reference to FIG. The data sent from the host server 10 is divided by the disk array controller 12 into fixed length data such as data 0, 1, 2. Furthermore, the parity P is data 0, 1, 2
512B created by taking exclusive OR from
Data. These data 0, 1, 2, parity P
Are arranged on the disks 1 to 4 as shown in FIG. Here, regarding the data format of the data 0, the additional check code CC0 which is the feature of this embodiment is 512B.
Data, which is created based on LA21 and is written following LA21, and the check code RCC23 is 5
It is created and written based on 12B data, LA21 and CC0 (220). Similarly, for data 1 and data 2, additional check codes CC1 (221) and CC2 (222) are written as shown. Also,
As for the parity P, CC0 (220), CC1 (221), CC2 (2
22) are all written.

On the other hand, in the present embodiment, the processing at the time of reading is different from usual, and in accordance with the present embodiment, one of the disks 30 to 33 in which the target data is arranged and the parity data 37 are always read. An error is detected by checking whether the read additional check code 22 matches the additional check code at the corresponding position of the read data. In the case of the present embodiment, error detection is performed by the following judgment.

(1) If the check code 23 and the data 20 do not match, the data is an error. It is considered that this is because correct data was not written due to a failure of the data system during writing in the disc.

(2) The check code 23 and the data 20 match, but the position information 21 embedded in the read data and the position where the data 20 should be (the disk array control device issues a write command). If the specified position) does not match, the mismatched data is an error. This is a case where the destination of the data written in the wrong position is read due to the failure of the built-in control unit of the disk 13.

(3) If the additional check codes 22 do not match, the data of the parity group is first read, and two types of data reconstruction are performed on the assumption that each mismatched data is incorrect. The result is compared with the additional check code 22, and the inconsistent side is erroneous data. This is a case where the old data is being read because the data to be written was written in the wrong position due to the control unit failure. That is, taking the case of (2) of FIG. 2 as an example, when the data 0 and the parity P are read and the additional check code does not match, the data 0
Or parity P is in error. Therefore, the combination of the data 1 and the parity P and the combination of the data 2 and the parity P are checked for the coincidence of the additional check code to detect any error of the data 0 and the parity P. If data 0 is found to be an error, data 1, data 2 and parity P
Data 0 is recovered from.

(4) If there is no mismatch or mismatch with the check code, the data is correct and can be used as it is.

If a data error is detected by the above judgment, all the data that make up the array are read again, and the data lost due to the failure is recovered from that data. At the time of recovery, except for rebuilding the additional check code data from the read data, the normal RAI
The process is exactly the same as the D data recovery procedure.

In this embodiment, one parity disk is placed for three data disks, but it is possible to easily adapt to other RAID configurations by increasing or decreasing the number of fields of the check code 23. Is.

In the present embodiment, information such as CRC is used as the identifier because this facilitates the error recovery process as a RAID device. For example, even if the write total number in which the data part is written is held in the parity part, the same error detection capability can be obtained in RAID5. In this case, with a parity group configuration of three or less, it is difficult to know which data is correct, and it is necessary to add a check code.

[0037]

As described above, according to the present invention, it is possible to construct a RAID device having error detection capability and suppressing performance degradation while using a disk with low error detection capability in the disk control unit.

[Brief description of drawings]

FIG. 1 is a diagram showing a schematic configuration of a disk array device according to an embodiment of the present invention.

FIG. 2 is a diagram showing a data format to be written on the disc in the present embodiment.

FIG. 3 is a diagram showing a data arrangement in a disc in the present embodiment.

[Explanation of symbols]

12 Disk array controller 13 discs 20 data 21 Location information 34,35,36 data area 37 Parity data area 110 communication path 121 Host communication controller 122 Control Processor 123 cash memory section 124 Disk controller

Claims (4)

[Claims] 1. In a disk array device configured by using a plurality of disk devices, a plurality of data disks forming a parity group,
While writing each additional check code generated based on each data written to each data disk, to the parity disk that constitutes the parity group,
A disk array control method characterized by writing all additional check codes. 2. In a disk array device configured by using a plurality of disk devices, a plurality of data disks forming a parity group are
Each additional check code generated based on each data written to each data disk is written, and all the additional check codes are written to the parity disks forming the parity group. A disk array control method, wherein an error check is performed by reading the additional check code of the parity disk. 3. The disk array control method according to claim 2, wherein the plurality of data disks and the parity disk are
A disk array control method characterized in that position information of data to be written in the disk array is written respectively, and the position information is used for error checking of data written at an incorrect position. 4. The disk array control system according to claim 3, wherein a check code generated from each piece of data, each piece of position information, and each piece of additional check code is written on a data disk and a parity disk that form a parity group. A disk array control method characterized by writing at the end of the format.